docs/source/proto.html

<!DOCTYPE html>

<html>
  <head>
    <title>Protocol Documentation</title>
    <meta charset="UTF-8">
    <link rel="stylesheet" type="text/css" href="https://fonts.googleapis.com/css?family=Ubuntu:400,700,400italic"/>
    <style>
      body {
        width: 60em;
        margin: 1em auto;
        color: #222;
        font-family: "Ubuntu", sans-serif;
        padding-bottom: 4em;
      }

      h1 {
        font-weight: normal;
        border-bottom: 1px solid #aaa;
        padding-bottom: 0.5ex;
      }

      h2 {
        border-bottom: 1px solid #aaa;
        padding-bottom: 0.5ex;
        margin: 1.5em 0;
      }

      h3 {
        font-weight: normal;
        border-bottom: 1px solid #aaa;
        padding-bottom: 0.5ex;
      }

      a {
        text-decoration: none;
        color: #567e25;
      }

      table {
        width: 100%;
        font-size: 80%;
        border-collapse: collapse;
      }

      thead {
        font-weight: 700;
        background-color: #dcdcdc;
      }

      tbody tr:nth-child(even) {
        background-color: #fbfbfb;
      }

      td {
        border: 1px solid #ccc;
        padding: 0.5ex 2ex;
      }

      td p {
        text-indent: 1em;
        margin: 0;
      }

      td p:nth-child(1) {
        text-indent: 0;
      }


      .field-table td:nth-child(1) {
        width: 10em;
      }
      .field-table td:nth-child(2) {
        width: 10em;
      }
      .field-table td:nth-child(3) {
        width: 6em;
      }
      .field-table td:nth-child(4) {
        width: auto;
      }


      .extension-table td:nth-child(1) {
        width: 10em;
      }
      .extension-table td:nth-child(2) {
        width: 10em;
      }
      .extension-table td:nth-child(3) {
        width: 10em;
      }
      .extension-table td:nth-child(4) {
        width: 5em;
      }
      .extension-table td:nth-child(5) {
        width: auto;
      }


      .enum-table td:nth-child(1) {
        width: 10em;
      }
      .enum-table td:nth-child(2) {
        width: 10em;
      }
      .enum-table td:nth-child(3) {
        width: auto;
      }


      .scalar-value-types-table tr {
        height: 3em;
      }


      #toc-container ul {
        list-style-type: none;
        padding-left: 1em;
        line-height: 180%;
        margin: 0;
      }
      #toc > li > a {
        font-weight: bold;
      }


      .file-heading {
        width: 100%;
        display: table;
        border-bottom: 1px solid #aaa;
        margin: 4em 0 1.5em 0;
      }
      .file-heading h2 {
        border: none;
        display: table-cell;
      }
      .file-heading a {
        text-align: right;
        display: table-cell;
      }


      .badge {
        width: 1.6em;
        height: 1.6em;
        display: inline-block;

        line-height: 1.6em;
        text-align: center;
        font-weight: bold;
        font-size: 60%;

        color: #89ba48;
        background-color: #dff0c8;

        margin: 0.5ex 1em 0.5ex -1em;
        border: 1px solid #fbfbfb;
        border-radius: 1ex;
      }
    </style>


    <link rel="stylesheet" type="text/css" href="stylesheet.css"/>
  </head>

  <body>

    <h1 id="title">Protocol Documentation</h1>

    <h2>Table of Contents</h2>

    <div id="toc-container">
      <ul id="toc">


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fautoint.proto">easy_rec/python/protos/autoint.proto</a>
            <ul>

                <li>
                  <a href="#protos.AutoInt"><span class="badge">M</span>AutoInt</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fcmbf.proto">easy_rec/python/protos/cmbf.proto</a>
            <ul>

                <li>
                  <a href="#protos.CMBF"><span class="badge">M</span>CMBF</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fcollaborative_metric_learning.proto">easy_rec/python/protos/collaborative_metric_learning.proto</a>
            <ul>

                <li>
                  <a href="#protos.CoMetricLearningI2I"><span class="badge">M</span>CoMetricLearningI2I</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdataset.proto">easy_rec/python/protos/dataset.proto</a>
            <ul>

                <li>
                  <a href="#protos.DatasetConfig"><span class="badge">M</span>DatasetConfig</a>
                </li>

                <li>
                  <a href="#protos.DatasetConfig.Field"><span class="badge">M</span>DatasetConfig.Field</a>
                </li>

                <li>
                  <a href="#protos.HardNegativeSampler"><span class="badge">M</span>HardNegativeSampler</a>
                </li>

                <li>
                  <a href="#protos.HardNegativeSamplerV2"><span class="badge">M</span>HardNegativeSamplerV2</a>
                </li>

                <li>
                  <a href="#protos.NegativeSampler"><span class="badge">M</span>NegativeSampler</a>
                </li>

                <li>
                  <a href="#protos.NegativeSamplerInMemory"><span class="badge">M</span>NegativeSamplerInMemory</a>
                </li>

                <li>
                  <a href="#protos.NegativeSamplerV2"><span class="badge">M</span>NegativeSamplerV2</a>
                </li>


                <li>
                  <a href="#protos.DatasetConfig.FieldType"><span class="badge">E</span>DatasetConfig.FieldType</a>
                </li>

                <li>
                  <a href="#protos.DatasetConfig.InputType"><span class="badge">E</span>DatasetConfig.InputType</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdata_source.proto">easy_rec/python/protos/data_source.proto</a>
            <ul>

                <li>
                  <a href="#protos.BinaryDataInput"><span class="badge">M</span>BinaryDataInput</a>
                </li>

                <li>
                  <a href="#protos.DatahubServer"><span class="badge">M</span>DatahubServer</a>
                </li>

                <li>
                  <a href="#protos.KafkaServer"><span class="badge">M</span>KafkaServer</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdbmtl.proto">easy_rec/python/protos/dbmtl.proto</a>
            <ul>

                <li>
                  <a href="#protos.DBMTL"><span class="badge">M</span>DBMTL</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdcn.proto">easy_rec/python/protos/dcn.proto</a>
            <ul>

                <li>
                  <a href="#protos.CrossTower"><span class="badge">M</span>CrossTower</a>
                </li>

                <li>
                  <a href="#protos.DCN"><span class="badge">M</span>DCN</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdeepfm.proto">easy_rec/python/protos/deepfm.proto</a>
            <ul>

                <li>
                  <a href="#protos.DeepFM"><span class="badge">M</span>DeepFM</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdlrm.proto">easy_rec/python/protos/dlrm.proto</a>
            <ul>

                <li>
                  <a href="#protos.DLRM"><span class="badge">M</span>DLRM</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdnn.proto">easy_rec/python/protos/dnn.proto</a>
            <ul>

                <li>
                  <a href="#protos.DNN"><span class="badge">M</span>DNN</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdropoutnet.proto">easy_rec/python/protos/dropoutnet.proto</a>
            <ul>

                <li>
                  <a href="#protos.DropoutNet"><span class="badge">M</span>DropoutNet</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fdssm.proto">easy_rec/python/protos/dssm.proto</a>
            <ul>

                <li>
                  <a href="#protos.DSSM"><span class="badge">M</span>DSSM</a>
                </li>

                <li>
                  <a href="#protos.DSSMTower"><span class="badge">M</span>DSSMTower</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2feasy_rec_model.proto">easy_rec/python/protos/easy_rec_model.proto</a>
            <ul>

                <li>
                  <a href="#protos.DummyModel"><span class="badge">M</span>DummyModel</a>
                </li>

                <li>
                  <a href="#protos.EasyRecModel"><span class="badge">M</span>EasyRecModel</a>
                </li>

                <li>
                  <a href="#protos.KD"><span class="badge">M</span>KD</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fesmm.proto">easy_rec/python/protos/esmm.proto</a>
            <ul>

                <li>
                  <a href="#protos.ESMM"><span class="badge">M</span>ESMM</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2feval.proto">easy_rec/python/protos/eval.proto</a>
            <ul>

                <li>
                  <a href="#protos.AUC"><span class="badge">M</span>AUC</a>
                </li>

                <li>
                  <a href="#protos.Accuracy"><span class="badge">M</span>Accuracy</a>
                </li>

                <li>
                  <a href="#protos.AvgPrecisionAtTopK"><span class="badge">M</span>AvgPrecisionAtTopK</a>
                </li>

                <li>
                  <a href="#protos.EvalConfig"><span class="badge">M</span>EvalConfig</a>
                </li>

                <li>
                  <a href="#protos.EvalMetrics"><span class="badge">M</span>EvalMetrics</a>
                </li>

                <li>
                  <a href="#protos.GAUC"><span class="badge">M</span>GAUC</a>
                </li>

                <li>
                  <a href="#protos.Max_F1"><span class="badge">M</span>Max_F1</a>
                </li>

                <li>
                  <a href="#protos.MeanAbsoluteError"><span class="badge">M</span>MeanAbsoluteError</a>
                </li>

                <li>
                  <a href="#protos.MeanSquaredError"><span class="badge">M</span>MeanSquaredError</a>
                </li>

                <li>
                  <a href="#protos.Precision"><span class="badge">M</span>Precision</a>
                </li>

                <li>
                  <a href="#protos.Recall"><span class="badge">M</span>Recall</a>
                </li>

                <li>
                  <a href="#protos.RecallAtTopK"><span class="badge">M</span>RecallAtTopK</a>
                </li>

                <li>
                  <a href="#protos.RootMeanSquaredError"><span class="badge">M</span>RootMeanSquaredError</a>
                </li>

                <li>
                  <a href="#protos.SessionAUC"><span class="badge">M</span>SessionAUC</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fexport.proto">easy_rec/python/protos/export.proto</a>
            <ul>

                <li>
                  <a href="#protos.ExportConfig"><span class="badge">M</span>ExportConfig</a>
                </li>

                <li>
                  <a href="#protos.MultiValueFields"><span class="badge">M</span>MultiValueFields</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2ffeature_config.proto">easy_rec/python/protos/feature_config.proto</a>
            <ul>

                <li>
                  <a href="#protos.AttentionCombiner"><span class="badge">M</span>AttentionCombiner</a>
                </li>

                <li>
                  <a href="#protos.EVParams"><span class="badge">M</span>EVParams</a>
                </li>

                <li>
                  <a href="#protos.FeatureConfig"><span class="badge">M</span>FeatureConfig</a>
                </li>

                <li>
                  <a href="#protos.FeatureConfigV2"><span class="badge">M</span>FeatureConfigV2</a>
                </li>

                <li>
                  <a href="#protos.FeatureGroupConfig"><span class="badge">M</span>FeatureGroupConfig</a>
                </li>

                <li>
                  <a href="#protos.MultiHeadAttentionCombiner"><span class="badge">M</span>MultiHeadAttentionCombiner</a>
                </li>

                <li>
                  <a href="#protos.SeqAttGroupConfig"><span class="badge">M</span>SeqAttGroupConfig</a>
                </li>

                <li>
                  <a href="#protos.SeqAttMap"><span class="badge">M</span>SeqAttMap</a>
                </li>

                <li>
                  <a href="#protos.SequenceCombiner"><span class="badge">M</span>SequenceCombiner</a>
                </li>

                <li>
                  <a href="#protos.TextCnnCombiner"><span class="badge">M</span>TextCnnCombiner</a>
                </li>


                <li>
                  <a href="#protos.FeatureConfig.FeatureType"><span class="badge">E</span>FeatureConfig.FeatureType</a>
                </li>

                <li>
                  <a href="#protos.FeatureConfig.FieldType"><span class="badge">E</span>FeatureConfig.FieldType</a>
                </li>

                <li>
                  <a href="#protos.WideOrDeep"><span class="badge">E</span>WideOrDeep</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2ffm.proto">easy_rec/python/protos/fm.proto</a>
            <ul>

                <li>
                  <a href="#protos.FM"><span class="badge">M</span>FM</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fhive_config.proto">easy_rec/python/protos/hive_config.proto</a>
            <ul>

                <li>
                  <a href="#protos.HiveConfig"><span class="badge">M</span>HiveConfig</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fhyperparams.proto">easy_rec/python/protos/hyperparams.proto</a>
            <ul>

                <li>
                  <a href="#protos.ConstantInitializer"><span class="badge">M</span>ConstantInitializer</a>
                </li>

                <li>
                  <a href="#protos.GlorotNormalInitializer"><span class="badge">M</span>GlorotNormalInitializer</a>
                </li>

                <li>
                  <a href="#protos.Initializer"><span class="badge">M</span>Initializer</a>
                </li>

                <li>
                  <a href="#protos.L1L2Regularizer"><span class="badge">M</span>L1L2Regularizer</a>
                </li>

                <li>
                  <a href="#protos.L1Regularizer"><span class="badge">M</span>L1Regularizer</a>
                </li>

                <li>
                  <a href="#protos.L2Regularizer"><span class="badge">M</span>L2Regularizer</a>
                </li>

                <li>
                  <a href="#protos.RandomNormalInitializer"><span class="badge">M</span>RandomNormalInitializer</a>
                </li>

                <li>
                  <a href="#protos.Regularizer"><span class="badge">M</span>Regularizer</a>
                </li>

                <li>
                  <a href="#protos.TruncatedNormalInitializer"><span class="badge">M</span>TruncatedNormalInitializer</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2flayer.proto">easy_rec/python/protos/layer.proto</a>
            <ul>

                <li>
                  <a href="#protos.CMBFTower"><span class="badge">M</span>CMBFTower</a>
                </li>

                <li>
                  <a href="#protos.HighWayTower"><span class="badge">M</span>HighWayTower</a>
                </li>

                <li>
                  <a href="#protos.UniterTower"><span class="badge">M</span>UniterTower</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2floss.proto">easy_rec/python/protos/loss.proto</a>
            <ul>

                <li>
                  <a href="#protos.CircleLoss"><span class="badge">M</span>CircleLoss</a>
                </li>

                <li>
                  <a href="#protos.F1ReweighedLoss"><span class="badge">M</span>F1ReweighedLoss</a>
                </li>

                <li>
                  <a href="#protos.Loss"><span class="badge">M</span>Loss</a>
                </li>

                <li>
                  <a href="#protos.MultiSimilarityLoss"><span class="badge">M</span>MultiSimilarityLoss</a>
                </li>

                <li>
                  <a href="#protos.SoftmaxCrossEntropyWithNegativeMining"><span class="badge">M</span>SoftmaxCrossEntropyWithNegativeMining</a>
                </li>


                <li>
                  <a href="#protos.LossType"><span class="badge">E</span>LossType</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fmind.proto">easy_rec/python/protos/mind.proto</a>
            <ul>

                <li>
                  <a href="#protos.Capsule"><span class="badge">M</span>Capsule</a>
                </li>

                <li>
                  <a href="#protos.MIND"><span class="badge">M</span>MIND</a>
                </li>


                <li>
                  <a href="#protos.MIND.UserSeqCombineMethod"><span class="badge">E</span>MIND.UserSeqCombineMethod</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fmmoe.proto">easy_rec/python/protos/mmoe.proto</a>
            <ul>

                <li>
                  <a href="#protos.ExpertTower"><span class="badge">M</span>ExpertTower</a>
                </li>

                <li>
                  <a href="#protos.MMoE"><span class="badge">M</span>MMoE</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fmulti_tower.proto">easy_rec/python/protos/multi_tower.proto</a>
            <ul>

                <li>
                  <a href="#protos.BSTTower"><span class="badge">M</span>BSTTower</a>
                </li>

                <li>
                  <a href="#protos.DINTower"><span class="badge">M</span>DINTower</a>
                </li>

                <li>
                  <a href="#protos.MultiTower"><span class="badge">M</span>MultiTower</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fmulti_tower_recall.proto">easy_rec/python/protos/multi_tower_recall.proto</a>
            <ul>

                <li>
                  <a href="#protos.MultiTowerRecall"><span class="badge">M</span>MultiTowerRecall</a>
                </li>

                <li>
                  <a href="#protos.RecallTower"><span class="badge">M</span>RecallTower</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2foptimizer.proto">easy_rec/python/protos/optimizer.proto</a>
            <ul>

                <li>
                  <a href="#protos.AdagradOptimizer"><span class="badge">M</span>AdagradOptimizer</a>
                </li>

                <li>
                  <a href="#protos.AdamAsyncOptimizer"><span class="badge">M</span>AdamAsyncOptimizer</a>
                </li>

                <li>
                  <a href="#protos.AdamAsyncWOptimizer"><span class="badge">M</span>AdamAsyncWOptimizer</a>
                </li>

                <li>
                  <a href="#protos.AdamOptimizer"><span class="badge">M</span>AdamOptimizer</a>
                </li>

                <li>
                  <a href="#protos.AdamWOptimizer"><span class="badge">M</span>AdamWOptimizer</a>
                </li>

                <li>
                  <a href="#protos.ConstantLearningRate"><span class="badge">M</span>ConstantLearningRate</a>
                </li>

                <li>
                  <a href="#protos.CosineDecayLearningRate"><span class="badge">M</span>CosineDecayLearningRate</a>
                </li>

                <li>
                  <a href="#protos.ExponentialDecayLearningRate"><span class="badge">M</span>ExponentialDecayLearningRate</a>
                </li>

                <li>
                  <a href="#protos.FtrlOptimizer"><span class="badge">M</span>FtrlOptimizer</a>
                </li>

                <li>
                  <a href="#protos.LearningRate"><span class="badge">M</span>LearningRate</a>
                </li>

                <li>
                  <a href="#protos.ManualStepLearningRate"><span class="badge">M</span>ManualStepLearningRate</a>
                </li>

                <li>
                  <a href="#protos.ManualStepLearningRate.LearningRateSchedule"><span class="badge">M</span>ManualStepLearningRate.LearningRateSchedule</a>
                </li>

                <li>
                  <a href="#protos.MomentumOptimizer"><span class="badge">M</span>MomentumOptimizer</a>
                </li>

                <li>
                  <a href="#protos.MomentumWOptimizer"><span class="badge">M</span>MomentumWOptimizer</a>
                </li>

                <li>
                  <a href="#protos.Optimizer"><span class="badge">M</span>Optimizer</a>
                </li>

                <li>
                  <a href="#protos.PolyDecayLearningRate"><span class="badge">M</span>PolyDecayLearningRate</a>
                </li>

                <li>
                  <a href="#protos.RMSPropOptimizer"><span class="badge">M</span>RMSPropOptimizer</a>
                </li>

                <li>
                  <a href="#protos.TransformerLearningRate"><span class="badge">M</span>TransformerLearningRate</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fpipeline.proto">easy_rec/python/protos/pipeline.proto</a>
            <ul>

                <li>
                  <a href="#protos.EasyRecConfig"><span class="badge">M</span>EasyRecConfig</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fple.proto">easy_rec/python/protos/ple.proto</a>
            <ul>

                <li>
                  <a href="#protos.ExtractionNetwork"><span class="badge">M</span>ExtractionNetwork</a>
                </li>

                <li>
                  <a href="#protos.PLE"><span class="badge">M</span>PLE</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2frocket_launching.proto">easy_rec/python/protos/rocket_launching.proto</a>
            <ul>

                <li>
                  <a href="#protos.RocketLaunching"><span class="badge">M</span>RocketLaunching</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fsimi.proto">easy_rec/python/protos/simi.proto</a>
            <ul>


                <li>
                  <a href="#protos.Similarity"><span class="badge">E</span>Similarity</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fsimple_multi_task.proto">easy_rec/python/protos/simple_multi_task.proto</a>
            <ul>

                <li>
                  <a href="#protos.SimpleMultiTask"><span class="badge">M</span>SimpleMultiTask</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2ftf_predict.proto">easy_rec/python/protos/tf_predict.proto</a>
            <ul>

                <li>
                  <a href="#tensorflow.eas.ArrayProto"><span class="badge">M</span>ArrayProto</a>
                </li>

                <li>
                  <a href="#tensorflow.eas.ArrayShape"><span class="badge">M</span>ArrayShape</a>
                </li>

                <li>
                  <a href="#tensorflow.eas.PredictRequest"><span class="badge">M</span>PredictRequest</a>
                </li>

                <li>
                  <a href="#tensorflow.eas.PredictRequest.InputsEntry"><span class="badge">M</span>PredictRequest.InputsEntry</a>
                </li>

                <li>
                  <a href="#tensorflow.eas.PredictResponse"><span class="badge">M</span>PredictResponse</a>
                </li>

                <li>
                  <a href="#tensorflow.eas.PredictResponse.OutputsEntry"><span class="badge">M</span>PredictResponse.OutputsEntry</a>
                </li>


                <li>
                  <a href="#tensorflow.eas.ArrayDataType"><span class="badge">E</span>ArrayDataType</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2ftower.proto">easy_rec/python/protos/tower.proto</a>
            <ul>

                <li>
                  <a href="#protos.BayesTaskTower"><span class="badge">M</span>BayesTaskTower</a>
                </li>

                <li>
                  <a href="#protos.TaskTower"><span class="badge">M</span>TaskTower</a>
                </li>

                <li>
                  <a href="#protos.Tower"><span class="badge">M</span>Tower</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2ftrain.proto">easy_rec/python/protos/train.proto</a>
            <ul>

                <li>
                  <a href="#protos.IncrementSaveConfig"><span class="badge">M</span>IncrementSaveConfig</a>
                </li>

                <li>
                  <a href="#protos.IncrementSaveConfig.Datahub"><span class="badge">M</span>IncrementSaveConfig.Datahub</a>
                </li>

                <li>
                  <a href="#protos.IncrementSaveConfig.Datahub.Consumer"><span class="badge">M</span>IncrementSaveConfig.Datahub.Consumer</a>
                </li>

                <li>
                  <a href="#protos.IncrementSaveConfig.File"><span class="badge">M</span>IncrementSaveConfig.File</a>
                </li>

                <li>
                  <a href="#protos.IncrementSaveConfig.Kafka"><span class="badge">M</span>IncrementSaveConfig.Kafka</a>
                </li>

                <li>
                  <a href="#protos.IncrementSaveConfig.Kafka.Consumer"><span class="badge">M</span>IncrementSaveConfig.Kafka.Consumer</a>
                </li>

                <li>
                  <a href="#protos.TrainConfig"><span class="badge">M</span>TrainConfig</a>
                </li>


                <li>
                  <a href="#protos.DistributionStrategy"><span class="badge">E</span>DistributionStrategy</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2funiter.proto">easy_rec/python/protos/uniter.proto</a>
            <ul>

                <li>
                  <a href="#protos.Uniter"><span class="badge">M</span>Uniter</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fvariational_dropout.proto">easy_rec/python/protos/variational_dropout.proto</a>
            <ul>

                <li>
                  <a href="#protos.VariationalDropoutLayer"><span class="badge">M</span>VariationalDropoutLayer</a>
                </li>


            </ul>
          </li>


          <li>
            <a href="#easy_rec%2fpython%2fprotos%2fwide_and_deep.proto">easy_rec/python/protos/wide_and_deep.proto</a>
            <ul>

                <li>
                  <a href="#protos.WideAndDeep"><span class="badge">M</span>WideAndDeep</a>
                </li>


            </ul>
          </li>

        <li><a href="#scalar-value-types">Scalar Value Types</a></li>
      </ul>
    </div>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/autoint.proto">easy_rec/python/protos/autoint.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.AutoInt">AutoInt</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>multi_head_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of heads Default: 1</pre></td>
                </tr>

                <tr>
                  <td>multi_head_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The dimension of heads </pre></td>
                </tr>

                <tr>
                  <td>interacting_layer_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of interacting layers Default: 1</pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/cmbf.proto">easy_rec/python/protos/cmbf.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.CMBF">CMBF</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>config</td>
                  <td><a href="#protos.CMBFTower">CMBFTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>final_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/collaborative_metric_learning.proto">easy_rec/python/protos/collaborative_metric_learning.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.CoMetricLearningI2I">CoMetricLearningI2I</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>session_id</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>highway</td>
                  <td><a href="#protos.HighWayTower">HighWayTower</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>input</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>output_l2_normalized_emb</td>
                  <td><a href="#bool">bool</a></td>
                  <td>required</td>
                  <td><pre> Default: true</pre></td>
                </tr>

                <tr>
                  <td>sample_id</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>circle_loss</td>
                  <td><a href="#protos.CircleLoss">CircleLoss</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>multi_similarity_loss</td>
                  <td><a href="#protos.MultiSimilarityLoss">MultiSimilarityLoss</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>item_id</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/dataset.proto">easy_rec/python/protos/dataset.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DatasetConfig">DatasetConfig</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>batch_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>mini batch size to use for training and evaluation. Default: 32</pre></td>
                </tr>

                <tr>
                  <td>auto_expand_input_fields</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>set auto_expand_input_fields to true to
auto_expand field[1-21] to field1, field2, ..., field21 Default: false</pre></td>
                </tr>

                <tr>
                  <td>label_fields</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>label fields, normally only one field is used.
For multiple target models such as MMOE
multiple label_fields will be set. </pre></td>
                </tr>

                <tr>
                  <td>label_sep</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>label separator </pre></td>
                </tr>

                <tr>
                  <td>label_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>repeated</td>
                  <td><pre>label dimensions which need to be set when there
are labels have dimension &gt; 1 </pre></td>
                </tr>

                <tr>
                  <td>shuffle</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>whether to shuffle data Default: true</pre></td>
                </tr>

                <tr>
                  <td>shuffle_buffer_size</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>shufffle buffer for better performance, even shuffle buffer is set,
it is suggested to do full data shuffle before training
especially when the performance of models is not good. Default: 32</pre></td>
                </tr>

                <tr>
                  <td>num_epochs</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>The number of times a data source is read. If set to zero, the data source
will be reused indefinitely. Default: 0</pre></td>
                </tr>

                <tr>
                  <td>prefetch_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Number of decoded batches to prefetch. Default: 32</pre></td>
                </tr>

                <tr>
                  <td>shard</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>shard dataset to 1/num_workers in distribute mode
this param is not used anymore Default: false</pre></td>
                </tr>

                <tr>
                  <td>file_shard</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>shard by file, not by sample, valid only for CSVInput Default: false</pre></td>
                </tr>

                <tr>
                  <td>input_type</td>
                  <td><a href="#protos.DatasetConfig.InputType">DatasetConfig.InputType</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>separator</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>separator of column features, only used for CSVInput*
not used in OdpsInput*
binary separators are supported:
  CTRL&#43;A could be set as &#39;\001&#39;
  CTRL&#43;B could be set as &#39;\002&#39;
  CTRL&#43;C could be set as &#39;\003&#39;
for RTPInput and OdpsRTPInput it is usually set
to &#39;\002&#39; Default: ,</pre></td>
                </tr>

                <tr>
                  <td>num_parallel_calls</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>parallel preproces of raw data, avoid using too small
or too large numbers(suggested be to small than
number of the cores) Default: 8</pre></td>
                </tr>

                <tr>
                  <td>selected_cols</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>only used for OdpsInput/OdpsInputV2/OdpsRTPInput, comma separated
for RTPInput, selected_cols use indices as column names
 such as &#39;1,2,4&#39;, where 1,2 are label columns, and
 4 is the feature column, column 0,3 are not used, </pre></td>
                </tr>

                <tr>
                  <td>selected_col_types</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>selected col types, only used for OdpsInput/OdpsInputV2
to avoid error setting of data types </pre></td>
                </tr>

                <tr>
                  <td>input_fields</td>
                  <td><a href="#protos.DatasetConfig.Field">DatasetConfig.Field</a></td>
                  <td>repeated</td>
                  <td><pre>the input fields must be the same number and in the
same order as data in csv files or odps tables </pre></td>
                </tr>

                <tr>
                  <td>rtp_separator</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>for RTPInput only Default: ;</pre></td>
                </tr>

                <tr>
                  <td>ignore_error</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>ignore some data errors
it is not suggested to set this parameter Default: false</pre></td>
                </tr>

                <tr>
                  <td>pai_worker_queue</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>whether to use pai global shuffle queue, only for OdpsInput,
OdpsInputV2, OdpsRTPInputV2 Default: false</pre></td>
                </tr>

                <tr>
                  <td>pai_worker_slice_num</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 100</pre></td>
                </tr>

                <tr>
                  <td>chief_redundant</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>if true, one worker will duplicate the data of the chief node
and undertake the gradient computation of the chief node Default: false</pre></td>
                </tr>

                <tr>
                  <td>sample_weight</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>input field for sample weight </pre></td>
                </tr>

                <tr>
                  <td>data_compression_type</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>the compression type of tfrecord </pre></td>
                </tr>

                <tr>
                  <td>n_data_batch_tfrecord</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>n data for one feature in tfrecord </pre></td>
                </tr>

                <tr>
                  <td>with_header</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>for csv files, may optionally with an header
in that case, input_name must match header name,
and the number and the order of input_fields
may not be the same as that in csv files. Default: false</pre></td>
                </tr>

                <tr>
                  <td>feature_fields</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>negative_sampler</td>
                  <td><a href="#protos.NegativeSampler">NegativeSampler</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>negative_sampler_v2</td>
                  <td><a href="#protos.NegativeSamplerV2">NegativeSamplerV2</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>hard_negative_sampler</td>
                  <td><a href="#protos.HardNegativeSampler">HardNegativeSampler</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>hard_negative_sampler_v2</td>
                  <td><a href="#protos.HardNegativeSamplerV2">HardNegativeSamplerV2</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>negative_sampler_in_memory</td>
                  <td><a href="#protos.NegativeSamplerInMemory">NegativeSamplerInMemory</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>eval_batch_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 4096</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.DatasetConfig.Field">DatasetConfig.Field</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>input_type</td>
                  <td><a href="#protos.DatasetConfig.FieldType">DatasetConfig.FieldType</a></td>
                  <td>required</td>
                  <td><pre> Default: STRING</pre></td>
                </tr>

                <tr>
                  <td>default_val</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>input_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>input_shape</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.HardNegativeSampler">HardNegativeSampler</h3>
        <pre>Weighted Random Sampling ItemID not in Batch and Sampling Hard Edge</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>user_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>user data path
userid weight </pre></td>
                </tr>

                <tr>
                  <td>item_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>item data path
itemid weight attrs </pre></td>
                </tr>

                <tr>
                  <td>hard_neg_edge_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>hard negative edge path
userid itemid weight </pre></td>
                </tr>

                <tr>
                  <td>num_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>number of negative sample </pre></td>
                </tr>

                <tr>
                  <td>num_hard_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>max number of hard negative sample </pre></td>
                </tr>

                <tr>
                  <td>attr_fields</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>field names of attrs in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>item_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of item_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>user_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of user_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>attr_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> Default: :</pre></td>
                </tr>

                <tr>
                  <td>num_eval_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>field_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>only works on DataScience/Local Default: </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.HardNegativeSamplerV2">HardNegativeSamplerV2</h3>
        <pre>Weighted Random Sampling ItemID not with Edge and Sampling Hard Edge</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>user_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>user data path
userid weight </pre></td>
                </tr>

                <tr>
                  <td>item_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>item data path
itemid weight attrs </pre></td>
                </tr>

                <tr>
                  <td>pos_edge_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>positive edge path
userid itemid weight </pre></td>
                </tr>

                <tr>
                  <td>hard_neg_edge_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>hard negative edge path
userid itemid weight </pre></td>
                </tr>

                <tr>
                  <td>num_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>number of negative sample </pre></td>
                </tr>

                <tr>
                  <td>num_hard_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>max number of hard negative sample </pre></td>
                </tr>

                <tr>
                  <td>attr_fields</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>field names of attrs in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>item_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of item_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>user_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of user_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>attr_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> Default: :</pre></td>
                </tr>

                <tr>
                  <td>num_eval_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>field_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>only works on DataScience/Local Default: </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.NegativeSampler">NegativeSampler</h3>
        <pre>Weighted Random Sampling ItemID not in Batch</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>sample data path
itemid weight attrs </pre></td>
                </tr>

                <tr>
                  <td>num_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>number of negative sample </pre></td>
                </tr>

                <tr>
                  <td>attr_fields</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>field names of attrs in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>item_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of item_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>attr_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> Default: :</pre></td>
                </tr>

                <tr>
                  <td>num_eval_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>field_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>only works on DataScience/Local Default: </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.NegativeSamplerInMemory">NegativeSamplerInMemory</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>sample data path
itemid weight attrs </pre></td>
                </tr>

                <tr>
                  <td>num_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>number of negative sample </pre></td>
                </tr>

                <tr>
                  <td>attr_fields</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>field names of attrs in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>item_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of item_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>attr_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> Default: :</pre></td>
                </tr>

                <tr>
                  <td>num_eval_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>field_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>only works on DataScience/Local Default: </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.NegativeSamplerV2">NegativeSamplerV2</h3>
        <pre>Weighted Random Sampling ItemID not with Edge</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>user_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>user data path
userid weight </pre></td>
                </tr>

                <tr>
                  <td>item_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>item data path
itemid weight attrs </pre></td>
                </tr>

                <tr>
                  <td>pos_edge_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>positive edge path
userid itemid weight </pre></td>
                </tr>

                <tr>
                  <td>num_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>number of negative sample </pre></td>
                </tr>

                <tr>
                  <td>attr_fields</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>field names of attrs in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>item_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of item_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>user_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>field name of user_id in train data or eval data </pre></td>
                </tr>

                <tr>
                  <td>attr_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> Default: :</pre></td>
                </tr>

                <tr>
                  <td>num_eval_sample</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>field_delimiter</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>only works on DataScience/Local Default: </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.DatasetConfig.FieldType">DatasetConfig.FieldType</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>INT32</td>
                <td>0</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>INT64</td>
                <td>1</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>STRING</td>
                <td>2</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>FLOAT</td>
                <td>4</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DOUBLE</td>
                <td>5</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>BOOL</td>
                <td>6</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>

        <h3 id="protos.DatasetConfig.InputType">DatasetConfig.InputType</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>CSVInput</td>
                <td>10</td>
                <td><pre>csv format input, could be used in local or hdfs
support .gz compression(but not .tar.gz files)</pre></td>
              </tr>

              <tr>
                <td>CSVInputV2</td>
                <td>11</td>
                <td><pre>@Depreciated</pre></td>
              </tr>

              <tr>
                <td>CSVInputEx</td>
                <td>12</td>
                <td><pre>extended csv format, allow quote in fields</pre></td>
              </tr>

              <tr>
                <td>OdpsInput</td>
                <td>2</td>
                <td><pre>@Depreciated, has memory leak problem</pre></td>
              </tr>

              <tr>
                <td>OdpsInputV2</td>
                <td>3</td>
                <td><pre>odps input, used on pai</pre></td>
              </tr>

              <tr>
                <td>DataHubInput</td>
                <td>15</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>OdpsInputV3</td>
                <td>9</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>RTPInput</td>
                <td>4</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>RTPInputV2</td>
                <td>5</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>OdpsRTPInput</td>
                <td>601</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>OdpsRTPInputV2</td>
                <td>602</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>TFRecordInput</td>
                <td>7</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>BatchTFRecordInput</td>
                <td>14</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DummyInput</td>
                <td>8</td>
                <td><pre>for the purpose to debug performance bottleneck of
input pipelines</pre></td>
              </tr>

              <tr>
                <td>KafkaInput</td>
                <td>13</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>HiveInput</td>
                <td>16</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>HiveRTPInput</td>
                <td>17</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>HiveParquetInput</td>
                <td>18</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>CriteoInput</td>
                <td>1001</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/data_source.proto">easy_rec/python/protos/data_source.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.BinaryDataInput">BinaryDataInput</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>category_path</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>support gfile.Glob </pre></td>
                </tr>

                <tr>
                  <td>dense_path</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>label_path</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.DatahubServer">DatahubServer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>akId</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>akSecret</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>endpoint</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>project</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>topic</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>offset_info</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>in json format: {&#34;0&#34;:{&#34;cursor&#34;: &#34;&#34;}, &#34;1&#34;:{&#34;cursor&#34;:&#34;&#34;}} </pre></td>
                </tr>

                <tr>
                  <td>offset_time</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>offset_time could be two formats:
1: %Y%m%d %H:%M:%S  &#34;20220508 12:00:00&#34;
2: %s               &#34;1651982400&#34; </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.KafkaServer">KafkaServer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>server</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>topic</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>group</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>offset_info</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>in json format: {&#39;0&#39;:10, &#39;1&#39;:20} </pre></td>
                </tr>

                <tr>
                  <td>offset_time</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>offset_time could be two formats:
1: %Y%m%d %H:%M:%S  &#39;20220508 12:00:00&#39;
2: %s               &#39;1651982400&#39; </pre></td>
                </tr>

                <tr>
                  <td>config_global</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>kafka global config, such as: fetch.max.bytes=1024 </pre></td>
                </tr>

                <tr>
                  <td>config_topic</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>kafka topic config, such as: max.partition.fetch.bytes=1024 </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/dbmtl.proto">easy_rec/python/protos/dbmtl.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DBMTL">DBMTL</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>bottom_cmbf</td>
                  <td><a href="#protos.CMBFTower">CMBFTower</a></td>
                  <td>optional</td>
                  <td><pre>shared bottom cmbf layer </pre></td>
                </tr>

                <tr>
                  <td>bottom_uniter</td>
                  <td><a href="#protos.UniterTower">UniterTower</a></td>
                  <td>optional</td>
                  <td><pre>shared bottom uniter layer </pre></td>
                </tr>

                <tr>
                  <td>bottom_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>shared bottom dnn layer </pre></td>
                </tr>

                <tr>
                  <td>expert_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>mmoe expert dnn layer definition </pre></td>
                </tr>

                <tr>
                  <td>num_expert</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>number of mmoe experts Default: 0</pre></td>
                </tr>

                <tr>
                  <td>task_towers</td>
                  <td><a href="#protos.BayesTaskTower">BayesTaskTower</a></td>
                  <td>repeated</td>
                  <td><pre>bayes task tower </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>l2 regularization Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/dcn.proto">easy_rec/python/protos/dcn.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.CrossTower">CrossTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>cross_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of cross layers Default: 3</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.DCN">DCN</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>deep_tower</td>
                  <td><a href="#protos.Tower">Tower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>cross_tower</td>
                  <td><a href="#protos.CrossTower">CrossTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>final_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/deepfm.proto">easy_rec/python/protos/deepfm.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DeepFM">DeepFM</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>final_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>wide_output_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>wide_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>deprecated Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>dense_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>deprecated Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/dlrm.proto">easy_rec/python/protos/dlrm.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DLRM">DLRM</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>top_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>bot_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>arch_interaction_op</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>options are: dot and cat Default: dot</pre></td>
                </tr>

                <tr>
                  <td>arch_interaction_itself</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>whether a feature will interact with itself Default: false</pre></td>
                </tr>

                <tr>
                  <td>arch_with_dense_feature</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>whether to include dense features after interaction Default: false</pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1e-05</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/dnn.proto">easy_rec/python/protos/dnn.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DNN">DNN</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>hidden_units</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>repeated</td>
                  <td><pre>hidden units for each layer </pre></td>
                </tr>

                <tr>
                  <td>dropout_ratio</td>
                  <td><a href="#float">float</a></td>
                  <td>repeated</td>
                  <td><pre>ratio of dropout </pre></td>
                </tr>

                <tr>
                  <td>activation</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>activation function Default: tf.nn.relu</pre></td>
                </tr>

                <tr>
                  <td>use_bn</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>use batch normalization Default: true</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/dropoutnet.proto">easy_rec/python/protos/dropoutnet.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DropoutNet">DropoutNet</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>user_content</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>user_preference</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>item_content</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>item_preference</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>user_tower</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>item_tower</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>user_dropout_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>item_dropout_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.5</pre></td>
                </tr>

                <tr>
                  <td>softmax_loss</td>
                  <td><a href="#protos.SoftmaxCrossEntropyWithNegativeMining">SoftmaxCrossEntropyWithNegativeMining</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/dssm.proto">easy_rec/python/protos/dssm.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DSSM">DSSM</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>user_tower</td>
                  <td><a href="#protos.DSSMTower">DSSMTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>item_tower</td>
                  <td><a href="#protos.DSSMTower">DSSMTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>simi_func</td>
                  <td><a href="#protos.Similarity">Similarity</a></td>
                  <td>optional</td>
                  <td><pre> Default: COSINE</pre></td>
                </tr>

                <tr>
                  <td>scale_simi</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>add a layer for scaling the similarity Default: true</pre></td>
                </tr>

                <tr>
                  <td>item_id</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>ignore_in_batch_neg_sam</td>
                  <td><a href="#bool">bool</a></td>
                  <td>required</td>
                  <td><pre> Default: false</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.DSSMTower">DSSMTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>id</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/easy_rec_model.proto">easy_rec/python/protos/easy_rec_model.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.DummyModel">DummyModel</h3>
        <pre>for input performance test</pre>


        <h3 id="protos.EasyRecModel">EasyRecModel</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>model_class</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>feature_groups</td>
                  <td><a href="#protos.FeatureGroupConfig">FeatureGroupConfig</a></td>
                  <td>repeated</td>
                  <td><pre>actually input layers, each layer produce a group of feature </pre></td>
                </tr>

                <tr>
                  <td>dummy</td>
                  <td><a href="#protos.DummyModel">DummyModel</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>wide_and_deep</td>
                  <td><a href="#protos.WideAndDeep">WideAndDeep</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>deepfm</td>
                  <td><a href="#protos.DeepFM">DeepFM</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>multi_tower</td>
                  <td><a href="#protos.MultiTower">MultiTower</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>fm</td>
                  <td><a href="#protos.FM">FM</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dcn</td>
                  <td><a href="#protos.DCN">DCN</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>autoint</td>
                  <td><a href="#protos.AutoInt">AutoInt</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dlrm</td>
                  <td><a href="#protos.DLRM">DLRM</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>cmbf</td>
                  <td><a href="#protos.CMBF">CMBF</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>uniter</td>
                  <td><a href="#protos.Uniter">Uniter</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>multi_tower_recall</td>
                  <td><a href="#protos.MultiTowerRecall">MultiTowerRecall</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dssm</td>
                  <td><a href="#protos.DSSM">DSSM</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>mind</td>
                  <td><a href="#protos.MIND">MIND</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dropoutnet</td>
                  <td><a href="#protos.DropoutNet">DropoutNet</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>metric_learning</td>
                  <td><a href="#protos.CoMetricLearningI2I">CoMetricLearningI2I</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>mmoe</td>
                  <td><a href="#protos.MMoE">MMoE</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>esmm</td>
                  <td><a href="#protos.ESMM">ESMM</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dbmtl</td>
                  <td><a href="#protos.DBMTL">DBMTL</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>simple_multi_task</td>
                  <td><a href="#protos.SimpleMultiTask">SimpleMultiTask</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>ple</td>
                  <td><a href="#protos.PLE">PLE</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>rocket_launching</td>
                  <td><a href="#protos.RocketLaunching">RocketLaunching</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>seq_att_groups</td>
                  <td><a href="#protos.SeqAttGroupConfig">SeqAttGroupConfig</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>embedding_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>implemented in easy_rec/python/model/easy_rec_estimator
add regularization to all variables with &#34;embedding_weights:&#34;
in name Default: 0</pre></td>
                </tr>

                <tr>
                  <td>loss_type</td>
                  <td><a href="#protos.LossType">LossType</a></td>
                  <td>optional</td>
                  <td><pre> Default: CLASSIFICATION</pre></td>
                </tr>

                <tr>
                  <td>num_class</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>ev_params</td>
                  <td><a href="#protos.EVParams">EVParams</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>kd</td>
                  <td><a href="#protos.KD">KD</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>restore_filters</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>filter variables matching any pattern in restore_filters
common filters are Adam, Momentum, etc. </pre></td>
                </tr>

                <tr>
                  <td>variational_dropout</td>
                  <td><a href="#protos.VariationalDropoutLayer">VariationalDropoutLayer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>losses</td>
                  <td><a href="#protos.Loss">Loss</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.KD">KD</h3>
        <pre>for knowledge distillation</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>loss_name</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>pred_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>pred_is_logits</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>default to be logits Default: true</pre></td>
                </tr>

                <tr>
                  <td>soft_label_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>for CROSS_ENTROPY_LOSS, soft_label must be logits instead of probs </pre></td>
                </tr>

                <tr>
                  <td>label_is_logits</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>default to be logits Default: true</pre></td>
                </tr>

                <tr>
                  <td>loss_type</td>
                  <td><a href="#protos.LossType">LossType</a></td>
                  <td>required</td>
                  <td><pre>currently only support CROSS_ENTROPY_LOSS and L2_LOSS </pre></td>
                </tr>

                <tr>
                  <td>loss_weight</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>temperature</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>only for loss_type == CROSS_ENTROPY_LOSS Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/esmm.proto">easy_rec/python/protos/esmm.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.ESMM">ESMM</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>groups</td>
                  <td><a href="#protos.Tower">Tower</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>ctr_tower</td>
                  <td><a href="#protos.TaskTower">TaskTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>cvr_tower</td>
                  <td><a href="#protos.TaskTower">TaskTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/eval.proto">easy_rec/python/protos/eval.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.AUC">AUC</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>num_thresholds</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 200</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.Accuracy">Accuracy</h3>
        <pre></pre>


        <h3 id="protos.AvgPrecisionAtTopK">AvgPrecisionAtTopK</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>topk</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 5</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.EvalConfig">EvalConfig</h3>
        <pre>Message for configuring EasyRecModel evaluation jobs (eval.py).</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>num_examples</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Number of examples to process of evaluation. Default: 0</pre></td>
                </tr>

                <tr>
                  <td>eval_interval_secs</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>How often to run evaluation. Default: 300</pre></td>
                </tr>

                <tr>
                  <td>max_evals</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Maximum number of times to run evaluation. If set to 0, will run forever. Default: 0</pre></td>
                </tr>

                <tr>
                  <td>save_graph</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>Whether the TensorFlow graph used for evaluation should be saved to disk. Default: false</pre></td>
                </tr>

                <tr>
                  <td>metrics_set</td>
                  <td><a href="#protos.EvalMetrics">EvalMetrics</a></td>
                  <td>repeated</td>
                  <td><pre>Type of metrics to use for evaluation.
possible values: </pre></td>
                </tr>

                <tr>
                  <td>eval_online</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>Evaluation online with batch forward data of training Default: false</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.EvalMetrics">EvalMetrics</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>auc</td>
                  <td><a href="#protos.AUC">AUC</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>recall_at_topk</td>
                  <td><a href="#protos.RecallAtTopK">RecallAtTopK</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>mean_absolute_error</td>
                  <td><a href="#protos.MeanAbsoluteError">MeanAbsoluteError</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>mean_squared_error</td>
                  <td><a href="#protos.MeanSquaredError">MeanSquaredError</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>accuracy</td>
                  <td><a href="#protos.Accuracy">Accuracy</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>max_f1</td>
                  <td><a href="#protos.Max_F1">Max_F1</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>root_mean_squared_error</td>
                  <td><a href="#protos.RootMeanSquaredError">RootMeanSquaredError</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>gauc</td>
                  <td><a href="#protos.GAUC">GAUC</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>session_auc</td>
                  <td><a href="#protos.SessionAUC">SessionAUC</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>recall</td>
                  <td><a href="#protos.Recall">Recall</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>precision</td>
                  <td><a href="#protos.Precision">Precision</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>precision_at_topk</td>
                  <td><a href="#protos.AvgPrecisionAtTopK">AvgPrecisionAtTopK</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.GAUC">GAUC</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>uid_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>uid field name </pre></td>
                </tr>

                <tr>
                  <td>reduction</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>reduction method for auc of different users
* &#34;mean&#34;: simple mean of different users
* &#34;mean_by_sample_num&#34;: weighted mean with sample num of different users
* &#34;mean_by_positive_num&#34;: weighted mean with positive sample num of different users Default: mean</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.Max_F1">Max_F1</h3>
        <pre></pre>


        <h3 id="protos.MeanAbsoluteError">MeanAbsoluteError</h3>
        <pre></pre>


        <h3 id="protos.MeanSquaredError">MeanSquaredError</h3>
        <pre></pre>


        <h3 id="protos.Precision">Precision</h3>
        <pre></pre>


        <h3 id="protos.Recall">Recall</h3>
        <pre></pre>


        <h3 id="protos.RecallAtTopK">RecallAtTopK</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>topk</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 5</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.RootMeanSquaredError">RootMeanSquaredError</h3>
        <pre></pre>


        <h3 id="protos.SessionAUC">SessionAUC</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>session_id_field</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>session id field name </pre></td>
                </tr>

                <tr>
                  <td>reduction</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>reduction: reduction method for auc of different sessions
* &#34;mean&#34;: simple mean of different sessions
* &#34;mean_by_sample_num&#34;: weighted mean with sample num of different sessions
* &#34;mean_by_positive_num&#34;: weighted mean with positive sample num of different sessions Default: mean</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/export.proto">easy_rec/python/protos/export.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.ExportConfig">ExportConfig</h3>
        <pre>Message for configuring exporting models.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>batch_size</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>batch size used for exported model, -1 indicates batch_size is None
which is only supported by classification model right now, while
other models support static batch_size Default: -1</pre></td>
                </tr>

                <tr>
                  <td>exporter_type</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>type of exporter [final | latest | best | none] when train_and_evaluation
final: performs a single export in the end of training
latest: regularly exports the serving graph and checkpoints
best: export the best model according to best_exporter_metric
none: do not perform export Default: final</pre></td>
                </tr>

                <tr>
                  <td>best_exporter_metric</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>the metric used to determine the best checkpoint Default: auc</pre></td>
                </tr>

                <tr>
                  <td>metric_bigger</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>metric value the bigger the best Default: true</pre></td>
                </tr>

                <tr>
                  <td>enable_early_stop</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>enable early stop Default: false</pre></td>
                </tr>

                <tr>
                  <td>early_stop_func</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>custom early stop function, format:
   early_stop_func(eval_results, early_stop_params)
return True if should stop </pre></td>
                </tr>

                <tr>
                  <td>early_stop_params</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>custom early stop parameters </pre></td>
                </tr>

                <tr>
                  <td>max_check_steps</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>early stop max check steps Default: 10000</pre></td>
                </tr>

                <tr>
                  <td>multi_placeholder</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>each feature has a placeholder Default: true</pre></td>
                </tr>

                <tr>
                  <td>exports_to_keep</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>export to keep, only for exporter_type in [best, latest] Default: 1</pre></td>
                </tr>

                <tr>
                  <td>multi_value_fields</td>
                  <td><a href="#protos.MultiValueFields">MultiValueFields</a></td>
                  <td>optional</td>
                  <td><pre>multi value field list </pre></td>
                </tr>

                <tr>
                  <td>placeholder_named_by_input</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>is placeholder named by input Default: false</pre></td>
                </tr>

                <tr>
                  <td>filter_inputs</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>filter out inputs, only keep effective ones Default: true</pre></td>
                </tr>

                <tr>
                  <td>export_features</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>export the original feature values as string Default: false</pre></td>
                </tr>

                <tr>
                  <td>export_rtp_outputs</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>export the outputs required by RTP Default: false</pre></td>
                </tr>

                <tr>
                  <td>asset_files</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>export asset files </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MultiValueFields">MultiValueFields</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input_name</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/feature_config.proto">easy_rec/python/protos/feature_config.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.AttentionCombiner">AttentionCombiner</h3>
        <pre></pre>


        <h3 id="protos.EVParams">EVParams</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>filter_freq</td>
                  <td><a href="#uint64">uint64</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>steps_to_live</td>
                  <td><a href="#uint64">uint64</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.FeatureConfig">FeatureConfig</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>feature_name</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>input_names</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>input field names: must be included in DatasetConfig.input_fields </pre></td>
                </tr>

                <tr>
                  <td>feature_type</td>
                  <td><a href="#protos.FeatureConfig.FeatureType">FeatureConfig.FeatureType</a></td>
                  <td>required</td>
                  <td><pre> Default: IdFeature</pre></td>
                </tr>

                <tr>
                  <td>embedding_name</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>embedding_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>hash_bucket_size</td>
                  <td><a href="#uint64">uint64</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>num_buckets</td>
                  <td><a href="#uint64">uint64</a></td>
                  <td>optional</td>
                  <td><pre>for categorical_column_with_identity Default: 0</pre></td>
                </tr>

                <tr>
                  <td>boundaries</td>
                  <td><a href="#double">double</a></td>
                  <td>repeated</td>
                  <td><pre>only for raw features </pre></td>
                </tr>

                <tr>
                  <td>separator</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>separator with in features Default: |</pre></td>
                </tr>

                <tr>
                  <td>kv_separator</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>delimeter to separator key from value </pre></td>
                </tr>

                <tr>
                  <td>seq_multi_sep</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>delimeter to separate sequence multi-values </pre></td>
                </tr>

                <tr>
                  <td>max_seq_len</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>truncate sequence data to max_seq_len </pre></td>
                </tr>

                <tr>
                  <td>vocab_file</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>vocab_list</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>shared_names</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>many other field share this config </pre></td>
                </tr>

                <tr>
                  <td>lookup_max_sel_elem_num</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>lookup max select element number, default 10 Default: 10</pre></td>
                </tr>

                <tr>
                  <td>max_partitions</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>max_partitions Default: 1</pre></td>
                </tr>

                <tr>
                  <td>combiner</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>combiner Default: sum</pre></td>
                </tr>

                <tr>
                  <td>initializer</td>
                  <td><a href="#protos.Initializer">Initializer</a></td>
                  <td>optional</td>
                  <td><pre>embedding initializer </pre></td>
                </tr>

                <tr>
                  <td>precision</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>number of digits kept after dot in format float/double to string
scientific format is not used.
in default it is not allowed to convert float/double to string Default: -1</pre></td>
                </tr>

                <tr>
                  <td>min_val</td>
                  <td><a href="#double">double</a></td>
                  <td>optional</td>
                  <td><pre>normalize raw feature to [0-1] Default: 0</pre></td>
                </tr>

                <tr>
                  <td>max_val</td>
                  <td><a href="#double">double</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>normalizer_fn</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>normalization function for raw features:
  such as: tf.math.log1p </pre></td>
                </tr>

                <tr>
                  <td>raw_input_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>raw feature of multiple dimensions Default: 1</pre></td>
                </tr>

                <tr>
                  <td>sequence_combiner</td>
                  <td><a href="#protos.SequenceCombiner">SequenceCombiner</a></td>
                  <td>optional</td>
                  <td><pre>sequence feature combiner </pre></td>
                </tr>

                <tr>
                  <td>sub_feature_type</td>
                  <td><a href="#protos.FeatureConfig.FeatureType">FeatureConfig.FeatureType</a></td>
                  <td>optional</td>
                  <td><pre>sub feature type for sequence feature Default: IdFeature</pre></td>
                </tr>

                <tr>
                  <td>sequence_length</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>sequence length Default: 1</pre></td>
                </tr>

                <tr>
                  <td>expression</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>for expr feature </pre></td>
                </tr>

                <tr>
                  <td>ev_params</td>
                  <td><a href="#protos.EVParams">EVParams</a></td>
                  <td>optional</td>
                  <td><pre>embedding variable params </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.FeatureConfigV2">FeatureConfigV2</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>features</td>
                  <td><a href="#protos.FeatureConfig">FeatureConfig</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.FeatureGroupConfig">FeatureGroupConfig</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>group_name</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>feature_names</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>wide_deep</td>
                  <td><a href="#protos.WideOrDeep">WideOrDeep</a></td>
                  <td>optional</td>
                  <td><pre> Default: DEEP</pre></td>
                </tr>

                <tr>
                  <td>sequence_features</td>
                  <td><a href="#protos.SeqAttGroupConfig">SeqAttGroupConfig</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>negative_sampler</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: false</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MultiHeadAttentionCombiner">MultiHeadAttentionCombiner</h3>
        <pre></pre>


        <h3 id="protos.SeqAttGroupConfig">SeqAttGroupConfig</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>group_name</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>seq_att_map</td>
                  <td><a href="#protos.SeqAttMap">SeqAttMap</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>tf_summary</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: false</pre></td>
                </tr>

                <tr>
                  <td>seq_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>allow_key_search</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: false</pre></td>
                </tr>

                <tr>
                  <td>need_key_feature</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: true</pre></td>
                </tr>

                <tr>
                  <td>allow_key_transform</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: false</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.SeqAttMap">SeqAttMap</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>key</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>hist_seq</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>aux_hist_seq</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.SequenceCombiner">SequenceCombiner</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>attention</td>
                  <td><a href="#protos.AttentionCombiner">AttentionCombiner</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>multi_head_attention</td>
                  <td><a href="#protos.MultiHeadAttentionCombiner">MultiHeadAttentionCombiner</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>text_cnn</td>
                  <td><a href="#protos.TextCnnCombiner">TextCnnCombiner</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.TextCnnCombiner">TextCnnCombiner</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>filter_sizes</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>num_filters</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.FeatureConfig.FeatureType">FeatureConfig.FeatureType</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>IdFeature</td>
                <td>0</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>RawFeature</td>
                <td>1</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>TagFeature</td>
                <td>2</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>ComboFeature</td>
                <td>3</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>LookupFeature</td>
                <td>4</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>SequenceFeature</td>
                <td>5</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>ExprFeature</td>
                <td>6</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>

        <h3 id="protos.FeatureConfig.FieldType">FeatureConfig.FieldType</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>INT32</td>
                <td>0</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>INT64</td>
                <td>1</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>STRING</td>
                <td>2</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>FLOAT</td>
                <td>4</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DOUBLE</td>
                <td>5</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>BOOL</td>
                <td>6</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>

        <h3 id="protos.WideOrDeep">WideOrDeep</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>DEEP</td>
                <td>0</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>WIDE</td>
                <td>1</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>WIDE_AND_DEEP</td>
                <td>2</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/fm.proto">easy_rec/python/protos/fm.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.FM">FM</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/hive_config.proto">easy_rec/python/protos/hive_config.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.HiveConfig">HiveConfig</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>host</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>hive master&#39;s ip </pre></td>
                </tr>

                <tr>
                  <td>port</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>hive port Default: 10000</pre></td>
                </tr>

                <tr>
                  <td>username</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>hive username Default: admin</pre></td>
                </tr>

                <tr>
                  <td>database</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>hive database Default: default</pre></td>
                </tr>

                <tr>
                  <td>table_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/hyperparams.proto">easy_rec/python/protos/hyperparams.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.ConstantInitializer">ConstantInitializer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>consts</td>
                  <td><a href="#float">float</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.GlorotNormalInitializer">GlorotNormalInitializer</h3>
        <pre></pre>


        <h3 id="protos.Initializer">Initializer</h3>
        <pre>Proto with one-of field for initializers.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>truncated_normal_initializer</td>
                  <td><a href="#protos.TruncatedNormalInitializer">TruncatedNormalInitializer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>random_normal_initializer</td>
                  <td><a href="#protos.RandomNormalInitializer">RandomNormalInitializer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>glorot_normal_initializer</td>
                  <td><a href="#protos.GlorotNormalInitializer">GlorotNormalInitializer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>constant_initializer</td>
                  <td><a href="#protos.ConstantInitializer">ConstantInitializer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.L1L2Regularizer">L1L2Regularizer</h3>
        <pre>Configuration proto for L2 Regularizer.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>scale_l1</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>scale_l2</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.L1Regularizer">L1Regularizer</h3>
        <pre>Configuration proto for L1 Regularizer.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>scale</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.L2Regularizer">L2Regularizer</h3>
        <pre>Configuration proto for L2 Regularizer.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>scale</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.RandomNormalInitializer">RandomNormalInitializer</h3>
        <pre>Configuration proto for random normal initializer. See</pre><pre>https://www.tensorflow.org/api_docs/python/tf/random_normal_initializer</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>mean</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>stddev</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.Regularizer">Regularizer</h3>
        <pre>Proto with one-of field for regularizers.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>l1_regularizer</td>
                  <td><a href="#protos.L1Regularizer">L1Regularizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularizer</td>
                  <td><a href="#protos.L2Regularizer">L2Regularizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l1_l2_regularizer</td>
                  <td><a href="#protos.L1L2Regularizer">L1L2Regularizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.TruncatedNormalInitializer">TruncatedNormalInitializer</h3>
        <pre>Configuration proto for truncated normal initializer. See</pre><pre>https://www.tensorflow.org/api_docs/python/tf/truncated_normal_initializer</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>mean</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>stddev</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/layer.proto">easy_rec/python/protos/layer.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.CMBFTower">CMBFTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>multi_head_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of heads of cross modal fusion layer Default: 1</pre></td>
                </tr>

                <tr>
                  <td>image_multi_head_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of heads of image feature learning layer Default: 1</pre></td>
                </tr>

                <tr>
                  <td>text_multi_head_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of heads of text feature learning layer Default: 1</pre></td>
                </tr>

                <tr>
                  <td>text_head_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The dimension of text heads </pre></td>
                </tr>

                <tr>
                  <td>image_head_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The dimension of image heads Default: 64</pre></td>
                </tr>

                <tr>
                  <td>image_feature_patch_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of patches of image feature, take effect when there is only one image feature Default: 1</pre></td>
                </tr>

                <tr>
                  <td>image_feature_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>Do dimension reduce to this size for image feature before single modal learning module Default: 0</pre></td>
                </tr>

                <tr>
                  <td>image_self_attention_layer_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of self attention layers for image features Default: 0</pre></td>
                </tr>

                <tr>
                  <td>text_self_attention_layer_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of self attention layers for text features Default: 1</pre></td>
                </tr>

                <tr>
                  <td>cross_modal_layer_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The number of cross modal layers Default: 1</pre></td>
                </tr>

                <tr>
                  <td>image_cross_head_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The dimension of image cross modal heads </pre></td>
                </tr>

                <tr>
                  <td>text_cross_head_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The dimension of text cross modal heads </pre></td>
                </tr>

                <tr>
                  <td>hidden_dropout_prob</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre>Dropout probability for hidden layers Default: 0</pre></td>
                </tr>

                <tr>
                  <td>attention_probs_dropout_prob</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre>Dropout probability of the attention probabilities Default: 0</pre></td>
                </tr>

                <tr>
                  <td>use_token_type</td>
                  <td><a href="#bool">bool</a></td>
                  <td>required</td>
                  <td><pre>Whether to add embeddings for different text sequence features Default: false</pre></td>
                </tr>

                <tr>
                  <td>use_position_embeddings</td>
                  <td><a href="#bool">bool</a></td>
                  <td>required</td>
                  <td><pre>Whether to add position embeddings for the position of each token in the text sequence Default: true</pre></td>
                </tr>

                <tr>
                  <td>max_position_embeddings</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>Maximum sequence length that might ever be used with this model Default: 0</pre></td>
                </tr>

                <tr>
                  <td>text_seq_emb_dropout_prob</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre>Dropout probability for text sequence embeddings Default: 0.1</pre></td>
                </tr>

                <tr>
                  <td>other_feature_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>dnn layers for other features </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.HighWayTower">HighWayTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>emb_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.UniterTower">UniterTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>hidden_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>Size of the encoder layers and the pooler layer </pre></td>
                </tr>

                <tr>
                  <td>num_hidden_layers</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>Number of hidden layers in the Transformer encoder </pre></td>
                </tr>

                <tr>
                  <td>num_attention_heads</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>Number of attention heads for each attention layer in the Transformer encoder </pre></td>
                </tr>

                <tr>
                  <td>intermediate_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The size of the &#34;intermediate&#34; (i.e. feed-forward) layer in the Transformer encoder </pre></td>
                </tr>

                <tr>
                  <td>hidden_act</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>The non-linear activation function (function or string) in the encoder and pooler.

&#34;gelu&#34;, &#34;relu&#34;, &#34;tanh&#34; and &#34;swish&#34; are supported. Default: gelu</pre></td>
                </tr>

                <tr>
                  <td>hidden_dropout_prob</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre>The dropout probability for all fully connected layers in the embeddings, encoder, and pooler Default: 0.1</pre></td>
                </tr>

                <tr>
                  <td>attention_probs_dropout_prob</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre>The dropout ratio for the attention probabilities Default: 0.1</pre></td>
                </tr>

                <tr>
                  <td>max_position_embeddings</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>The maximum sequence length that this model might ever be used with Default: 512</pre></td>
                </tr>

                <tr>
                  <td>use_position_embeddings</td>
                  <td><a href="#bool">bool</a></td>
                  <td>required</td>
                  <td><pre>Whether to add position embeddings for the position of each token in the text sequence Default: true</pre></td>
                </tr>

                <tr>
                  <td>initializer_range</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre>The stddev of the truncated_normal_initializer for initializing all weight matrices Default: 0.02</pre></td>
                </tr>

                <tr>
                  <td>other_feature_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>dnn layers for other features </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/loss.proto">easy_rec/python/protos/loss.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.CircleLoss">CircleLoss</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>margin</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.25</pre></td>
                </tr>

                <tr>
                  <td>gamma</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 32</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.F1ReweighedLoss">F1ReweighedLoss</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>f1_beta_square</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>label_smoothing</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.Loss">Loss</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>loss_type</td>
                  <td><a href="#protos.LossType">LossType</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>weight</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>f1_reweighted_loss</td>
                  <td><a href="#protos.F1ReweighedLoss">F1ReweighedLoss</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>softmax_loss</td>
                  <td><a href="#protos.SoftmaxCrossEntropyWithNegativeMining">SoftmaxCrossEntropyWithNegativeMining</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>circle_loss</td>
                  <td><a href="#protos.CircleLoss">CircleLoss</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>multi_simi_loss</td>
                  <td><a href="#protos.MultiSimilarityLoss">MultiSimilarityLoss</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MultiSimilarityLoss">MultiSimilarityLoss</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>alpha</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 2</pre></td>
                </tr>

                <tr>
                  <td>beta</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 50</pre></td>
                </tr>

                <tr>
                  <td>lamb</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>eps</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.SoftmaxCrossEntropyWithNegativeMining">SoftmaxCrossEntropyWithNegativeMining</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>num_negative_samples</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>margin</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>gamma</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>coefficient_of_support_vector</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.LossType">LossType</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>CLASSIFICATION</td>
                <td>0</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>L2_LOSS</td>
                <td>1</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>SIGMOID_L2_LOSS</td>
                <td>2</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>CROSS_ENTROPY_LOSS</td>
                <td>3</td>
                <td><pre>crossentropy loss/log loss</pre></td>
              </tr>

              <tr>
                <td>SOFTMAX_CROSS_ENTROPY</td>
                <td>4</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>CIRCLE_LOSS</td>
                <td>5</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>MULTI_SIMILARITY_LOSS</td>
                <td>6</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>SOFTMAX_CROSS_ENTROPY_WITH_NEGATIVE_MINING</td>
                <td>7</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>PAIR_WISE_LOSS</td>
                <td>8</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>F1_REWEIGHTED_LOSS</td>
                <td>9</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/mind.proto">easy_rec/python/protos/mind.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.Capsule">Capsule</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>max_k</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>max number of high capsules Default: 5</pre></td>
                </tr>

                <tr>
                  <td>max_seq_len</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>max behaviour sequence length </pre></td>
                </tr>

                <tr>
                  <td>high_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>high capsule embedding vector dimension </pre></td>
                </tr>

                <tr>
                  <td>num_iters</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>number EM iterations Default: 3</pre></td>
                </tr>

                <tr>
                  <td>routing_logits_scale</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>routing logits scale Default: 20</pre></td>
                </tr>

                <tr>
                  <td>routing_logits_stddev</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>routing logits initial stddev Default: 1</pre></td>
                </tr>

                <tr>
                  <td>squash_pow</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>squash power Default: 1</pre></td>
                </tr>

                <tr>
                  <td>scale_ratio</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>output ratio Default: 1</pre></td>
                </tr>

                <tr>
                  <td>const_caps_num</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>constant interest number
in default, use log(seq_len) Default: false</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MIND">MIND</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>pre_capsule_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>preprocessing dnn before entering capsule layer </pre></td>
                </tr>

                <tr>
                  <td>user_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre>dnn layers applied on user_context(none sequence features) </pre></td>
                </tr>

                <tr>
                  <td>concat_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre>concat user and capsule dnn </pre></td>
                </tr>

                <tr>
                  <td>user_seq_combine</td>
                  <td><a href="#protos.MIND.UserSeqCombineMethod">MIND.UserSeqCombineMethod</a></td>
                  <td>optional</td>
                  <td><pre>method to combine several user sequences
such as item_ids, category_ids Default: SUM</pre></td>
                </tr>

                <tr>
                  <td>item_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre>dnn layers applied on item features </pre></td>
                </tr>

                <tr>
                  <td>capsule_config</td>
                  <td><a href="#protos.Capsule">Capsule</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>simi_pow</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>similarity power, the paper says that the big
the better Default: 10</pre></td>
                </tr>

                <tr>
                  <td>simi_func</td>
                  <td><a href="#protos.Similarity">Similarity</a></td>
                  <td>optional</td>
                  <td><pre> Default: COSINE</pre></td>
                </tr>

                <tr>
                  <td>scale_simi</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>add a layer for scaling the similarity Default: true</pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>time_id_fea</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>item_id</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>ignore_in_batch_neg_sam</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: false</pre></td>
                </tr>

                <tr>
                  <td>max_interests_simi</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>if small than 1.0, then a loss will be added to
limit the maximal interest similarities, but
in experiments, setup such a loss leads to low hitrate. Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MIND.UserSeqCombineMethod">MIND.UserSeqCombineMethod</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>CONCAT</td>
                <td>0</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>SUM</td>
                <td>1</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/mmoe.proto">easy_rec/python/protos/mmoe.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.ExpertTower">ExpertTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>expert_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MMoE">MMoE</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>experts</td>
                  <td><a href="#protos.ExpertTower">ExpertTower</a></td>
                  <td>repeated</td>
                  <td><pre>deprecated: original mmoe experts config </pre></td>
                </tr>

                <tr>
                  <td>expert_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>mmoe expert dnn layer definition </pre></td>
                </tr>

                <tr>
                  <td>num_expert</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>number of mmoe experts Default: 0</pre></td>
                </tr>

                <tr>
                  <td>task_towers</td>
                  <td><a href="#protos.TaskTower">TaskTower</a></td>
                  <td>repeated</td>
                  <td><pre>task tower </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre>l2 regularization Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/multi_tower.proto">easy_rec/python/protos/multi_tower.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.BSTTower">BSTTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>seq_len</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre> Default: 5</pre></td>
                </tr>

                <tr>
                  <td>multi_head_size</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre> Default: 4</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.DINTower">DINTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MultiTower">MultiTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>towers</td>
                  <td><a href="#protos.Tower">Tower</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>final_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>din_towers</td>
                  <td><a href="#protos.DINTower">DINTower</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>bst_towers</td>
                  <td><a href="#protos.BSTTower">BSTTower</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/multi_tower_recall.proto">easy_rec/python/protos/multi_tower_recall.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.MultiTowerRecall">MultiTowerRecall</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>user_tower</td>
                  <td><a href="#protos.RecallTower">RecallTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>item_tower</td>
                  <td><a href="#protos.RecallTower">RecallTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>final_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>ignore_in_batch_neg_sam</td>
                  <td><a href="#bool">bool</a></td>
                  <td>required</td>
                  <td><pre> Default: false</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.RecallTower">RecallTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/optimizer.proto">easy_rec/python/protos/optimizer.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.AdagradOptimizer">AdagradOptimizer</h3>
        <pre>Configuration message for the AdagradOptimizer</pre><pre>See: https://www.tensorflow.org/api_docs/python/tf/train/AdagradOptimizer</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.AdamAsyncOptimizer">AdamAsyncOptimizer</h3>
        <pre>Only available on pai-tf, which has better performance than AdamOptimizer</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>beta1</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

                <tr>
                  <td>beta2</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.999</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.AdamAsyncWOptimizer">AdamAsyncWOptimizer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>weight_decay</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1e-06</pre></td>
                </tr>

                <tr>
                  <td>beta1</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

                <tr>
                  <td>beta2</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.999</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.AdamOptimizer">AdamOptimizer</h3>
        <pre>Configuration message for the AdamOptimizer</pre><pre>See: https://www.tensorflow.org/api_docs/python/tf/train/AdamOptimizer</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>beta1</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

                <tr>
                  <td>beta2</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.999</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.AdamWOptimizer">AdamWOptimizer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>weight_decay</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1e-06</pre></td>
                </tr>

                <tr>
                  <td>beta1</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

                <tr>
                  <td>beta2</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.999</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.ConstantLearningRate">ConstantLearningRate</h3>
        <pre>Configuration message for a constant learning rate.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.002</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.CosineDecayLearningRate">CosineDecayLearningRate</h3>
        <pre>Configuration message for a cosine decaying learning rate as defined in</pre><pre>utils/learning_schedules.py</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate_base</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.002</pre></td>
                </tr>

                <tr>
                  <td>total_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 4000000</pre></td>
                </tr>

                <tr>
                  <td>warmup_learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.0002</pre></td>
                </tr>

                <tr>
                  <td>warmup_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 10000</pre></td>
                </tr>

                <tr>
                  <td>hold_base_rate_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.ExponentialDecayLearningRate">ExponentialDecayLearningRate</h3>
        <pre>Configuration message for an exponentially decaying learning rate.</pre><pre>See https://www.tensorflow.org/versions/master/api_docs/python/train/ \</pre><pre>decaying_the_learning_rate#exponential_decay</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>initial_learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.002</pre></td>
                </tr>

                <tr>
                  <td>decay_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 4000000</pre></td>
                </tr>

                <tr>
                  <td>decay_factor</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.95</pre></td>
                </tr>

                <tr>
                  <td>staircase</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: true</pre></td>
                </tr>

                <tr>
                  <td>burnin_learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>burnin_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>min_learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.FtrlOptimizer">FtrlOptimizer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre>optional float learning_rate = 1 [default=1e-4]; </pre></td>
                </tr>

                <tr>
                  <td>learning_rate_power</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: -0.5</pre></td>
                </tr>

                <tr>
                  <td>initial_accumulator_value</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.1</pre></td>
                </tr>

                <tr>
                  <td>l1_reg</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>l2_reg</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>l2_shrinkage_reg</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.LearningRate">LearningRate</h3>
        <pre>Configuration message for optimizer learning rate.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>constant_learning_rate</td>
                  <td><a href="#protos.ConstantLearningRate">ConstantLearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>exponential_decay_learning_rate</td>
                  <td><a href="#protos.ExponentialDecayLearningRate">ExponentialDecayLearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>manual_step_learning_rate</td>
                  <td><a href="#protos.ManualStepLearningRate">ManualStepLearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>cosine_decay_learning_rate</td>
                  <td><a href="#protos.CosineDecayLearningRate">CosineDecayLearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>poly_decay_learning_rate</td>
                  <td><a href="#protos.PolyDecayLearningRate">PolyDecayLearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>transformer_learning_rate</td>
                  <td><a href="#protos.TransformerLearningRate">TransformerLearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.ManualStepLearningRate">ManualStepLearningRate</h3>
        <pre>Configuration message for a manually defined learning rate schedule.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>initial_learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.002</pre></td>
                </tr>

                <tr>
                  <td>schedule</td>
                  <td><a href="#protos.ManualStepLearningRate.LearningRateSchedule">ManualStepLearningRate.LearningRateSchedule</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>warmup</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>Whether to linearly interpolate learning rates for steps in
[0, schedule[0].step]. Default: false</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.ManualStepLearningRate.LearningRateSchedule">ManualStepLearningRate.LearningRateSchedule</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>step</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.002</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MomentumOptimizer">MomentumOptimizer</h3>
        <pre>Configuration message for the MomentumOptimizer</pre><pre>See: https://www.tensorflow.org/api_docs/python/tf/train/MomentumOptimizer</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>momentum_optimizer_value</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.MomentumWOptimizer">MomentumWOptimizer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>weight_decay</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1e-06</pre></td>
                </tr>

                <tr>
                  <td>momentum_optimizer_value</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.Optimizer">Optimizer</h3>
        <pre>Top level optimizer message.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>rms_prop_optimizer</td>
                  <td><a href="#protos.RMSPropOptimizer">RMSPropOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>momentum_optimizer</td>
                  <td><a href="#protos.MomentumOptimizer">MomentumOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>adam_optimizer</td>
                  <td><a href="#protos.AdamOptimizer">AdamOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>momentumw_optimizer</td>
                  <td><a href="#protos.MomentumWOptimizer">MomentumWOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>adamw_optimizer</td>
                  <td><a href="#protos.AdamWOptimizer">AdamWOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>adam_async_optimizer</td>
                  <td><a href="#protos.AdamAsyncOptimizer">AdamAsyncOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>adagrad_optimizer</td>
                  <td><a href="#protos.AdagradOptimizer">AdagradOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>ftrl_optimizer</td>
                  <td><a href="#protos.FtrlOptimizer">FtrlOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>adam_asyncw_optimizer</td>
                  <td><a href="#protos.AdamAsyncWOptimizer">AdamAsyncWOptimizer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>use_moving_average</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: false</pre></td>
                </tr>

                <tr>
                  <td>moving_average_decay</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9999</pre></td>
                </tr>

                <tr>
                  <td>embedding_learning_rate_multiplier</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.PolyDecayLearningRate">PolyDecayLearningRate</h3>
        <pre>Configuration message for a poly decaying learning rate.</pre><pre>See https://www.tensorflow.org/api_docs/python/tf/train/polynomial_decay.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate_base</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>total_steps</td>
                  <td><a href="#int64">int64</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>power</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>end_learning_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.RMSPropOptimizer">RMSPropOptimizer</h3>
        <pre>Configuration message for the RMSPropOptimizer</pre><pre>See: https://www.tensorflow.org/api_docs/python/tf/train/RMSPropOptimizer</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate</td>
                  <td><a href="#protos.LearningRate">LearningRate</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>momentum_optimizer_value</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

                <tr>
                  <td>decay</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.9</pre></td>
                </tr>

                <tr>
                  <td>epsilon</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.TransformerLearningRate">TransformerLearningRate</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>learning_rate_base</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>hidden_size</td>
                  <td><a href="#int32">int32</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>warmup_steps</td>
                  <td><a href="#int32">int32</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>step_scaling_rate</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/pipeline.proto">easy_rec/python/protos/pipeline.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.EasyRecConfig">EasyRecConfig</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>train_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>kafka_train_input</td>
                  <td><a href="#protos.KafkaServer">KafkaServer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>datahub_train_input</td>
                  <td><a href="#protos.DatahubServer">DatahubServer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>hive_train_input</td>
                  <td><a href="#protos.HiveConfig">HiveConfig</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>binary_train_input</td>
                  <td><a href="#protos.BinaryDataInput">BinaryDataInput</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>eval_input_path</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>kafka_eval_input</td>
                  <td><a href="#protos.KafkaServer">KafkaServer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>datahub_eval_input</td>
                  <td><a href="#protos.DatahubServer">DatahubServer</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>hive_eval_input</td>
                  <td><a href="#protos.HiveConfig">HiveConfig</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>binary_eval_input</td>
                  <td><a href="#protos.BinaryDataInput">BinaryDataInput</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>model_dir</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>train_config</td>
                  <td><a href="#protos.TrainConfig">TrainConfig</a></td>
                  <td>optional</td>
                  <td><pre>train config, including optimizer, weight decay, num_steps and so on </pre></td>
                </tr>

                <tr>
                  <td>eval_config</td>
                  <td><a href="#protos.EvalConfig">EvalConfig</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>data_config</td>
                  <td><a href="#protos.DatasetConfig">DatasetConfig</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>feature_configs</td>
                  <td><a href="#protos.FeatureConfig">FeatureConfig</a></td>
                  <td>repeated</td>
                  <td><pre>for compatibility </pre></td>
                </tr>

                <tr>
                  <td>feature_config</td>
                  <td><a href="#protos.FeatureConfigV2">FeatureConfigV2</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>model_config</td>
                  <td><a href="#protos.EasyRecModel">EasyRecModel</a></td>
                  <td>required</td>
                  <td><pre>recommendation model config </pre></td>
                </tr>

                <tr>
                  <td>export_config</td>
                  <td><a href="#protos.ExportConfig">ExportConfig</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>fg_json_path</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>Json file[RTP FG] to define input data and features:
* In easy_rec.python.utils.fg_util.load_fg_json_to_config:
  data_config and feature_config will be generated
  based on fg_json.
* After generation, a prefix &#39;!&#39; is added:
  fg_json_path = &#39;!&#39; &#43; fg_json_path
  indicates config update is already done, and should not
  be updated anymore. In this way, we make load_fg_json_to_config
  function reentrant.
This step is done before edit_config_json to take effect. </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/ple.proto">easy_rec/python/protos/ple.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.ExtractionNetwork">ExtractionNetwork</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>network_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>expert_num_per_task</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre>number of experts per task </pre></td>
                </tr>

                <tr>
                  <td>share_num</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>number of experts for share
For the last extraction_network, no need to configure this </pre></td>
                </tr>

                <tr>
                  <td>task_expert_net</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre>dnn network of experts per task </pre></td>
                </tr>

                <tr>
                  <td>share_expert_net</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>dnn network of experts for share
For the last extraction_network, no need to configure this </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.PLE">PLE</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>extraction_networks</td>
                  <td><a href="#protos.ExtractionNetwork">ExtractionNetwork</a></td>
                  <td>repeated</td>
                  <td><pre>extraction network </pre></td>
                </tr>

                <tr>
                  <td>task_towers</td>
                  <td><a href="#protos.TaskTower">TaskTower</a></td>
                  <td>repeated</td>
                  <td><pre>task tower </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>l2 regularization Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/rocket_launching.proto">easy_rec/python/protos/rocket_launching.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.RocketLaunching">RocketLaunching</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>share_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>booster_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>light_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

                <tr>
                  <td>feature_based_distillation</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre> Default: false</pre></td>
                </tr>

                <tr>
                  <td>feature_distillation_function</td>
                  <td><a href="#protos.Similarity">Similarity</a></td>
                  <td>optional</td>
                  <td><pre>COSINE = 0; EUCLID = 1; Default: COSINE</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/simi.proto">easy_rec/python/protos/simi.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.Similarity">Similarity</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>COSINE</td>
                <td>0</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>INNER_PRODUCT</td>
                <td>1</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>EUCLID</td>
                <td>2</td>
                <td><pre></pre></td>
              </tr>

          </tbody>
        </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/simple_multi_task.proto">easy_rec/python/protos/simple_multi_task.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.SimpleMultiTask">SimpleMultiTask</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>task_towers</td>
                  <td><a href="#protos.TaskTower">TaskTower</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>required</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/tf_predict.proto">easy_rec/python/protos/tf_predict.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="tensorflow.eas.ArrayProto">ArrayProto</h3>
        <pre>Protocol buffer representing an array</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>dtype</td>
                  <td><a href="#tensorflow.eas.ArrayDataType">ArrayDataType</a></td>
                  <td></td>
                  <td><pre>Data Type. </pre></td>
                </tr>

                <tr>
                  <td>array_shape</td>
                  <td><a href="#tensorflow.eas.ArrayShape">ArrayShape</a></td>
                  <td></td>
                  <td><pre>Shape of the array. </pre></td>
                </tr>

                <tr>
                  <td>float_val</td>
                  <td><a href="#float">float</a></td>
                  <td>repeated</td>
                  <td><pre>DT_FLOAT. </pre></td>
                </tr>

                <tr>
                  <td>double_val</td>
                  <td><a href="#double">double</a></td>
                  <td>repeated</td>
                  <td><pre>DT_DOUBLE. </pre></td>
                </tr>

                <tr>
                  <td>int_val</td>
                  <td><a href="#int32">int32</a></td>
                  <td>repeated</td>
                  <td><pre>DT_INT32, DT_INT16, DT_INT8, DT_UINT8. </pre></td>
                </tr>

                <tr>
                  <td>string_val</td>
                  <td><a href="#bytes">bytes</a></td>
                  <td>repeated</td>
                  <td><pre>DT_STRING. </pre></td>
                </tr>

                <tr>
                  <td>int64_val</td>
                  <td><a href="#int64">int64</a></td>
                  <td>repeated</td>
                  <td><pre>DT_INT64. </pre></td>
                </tr>

                <tr>
                  <td>bool_val</td>
                  <td><a href="#bool">bool</a></td>
                  <td>repeated</td>
                  <td><pre>DT_BOOL. </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="tensorflow.eas.ArrayShape">ArrayShape</h3>
        <pre>Dimensions of an array</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>dim</td>
                  <td><a href="#int64">int64</a></td>
                  <td>repeated</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="tensorflow.eas.PredictRequest">PredictRequest</h3>
        <pre>PredictRequest specifies which TensorFlow model to run, as well as</pre><pre>how inputs are mapped to tensors and how outputs are filtered before</pre><pre>returning to user.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>signature_name</td>
                  <td><a href="#string">string</a></td>
                  <td></td>
                  <td><pre>A named signature to evaluate. If unspecified, the default signature
will be used </pre></td>
                </tr>

                <tr>
                  <td>inputs</td>
                  <td><a href="#tensorflow.eas.PredictRequest.InputsEntry">PredictRequest.InputsEntry</a></td>
                  <td>repeated</td>
                  <td><pre>Input tensors.
Names of input tensor are alias names. The mapping from aliases to real
input tensor names is expected to be stored as named generic signature
under the key &#34;inputs&#34; in the model export.
Each alias listed in a generic signature named &#34;inputs&#34; should be provided
exactly once in order to run the prediction. </pre></td>
                </tr>

                <tr>
                  <td>output_filter</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>Output filter.
Names specified are alias names. The mapping from aliases to real output
tensor names is expected to be stored as named generic signature under
the key &#34;outputs&#34; in the model export.
Only tensors specified here will be run/fetched and returned, with the
exception that when none is specified, all tensors specified in the
named signature will be run/fetched and returned. </pre></td>
                </tr>

                <tr>
                  <td>debug_level</td>
                  <td><a href="#int32">int32</a></td>
                  <td></td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="tensorflow.eas.PredictRequest.InputsEntry">PredictRequest.InputsEntry</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>key</td>
                  <td><a href="#string">string</a></td>
                  <td></td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>value</td>
                  <td><a href="#tensorflow.eas.ArrayProto">ArrayProto</a></td>
                  <td></td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="tensorflow.eas.PredictResponse">PredictResponse</h3>
        <pre>Response for PredictRequest on successful run.</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>outputs</td>
                  <td><a href="#tensorflow.eas.PredictResponse.OutputsEntry">PredictResponse.OutputsEntry</a></td>
                  <td>repeated</td>
                  <td><pre>Output tensors. </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="tensorflow.eas.PredictResponse.OutputsEntry">PredictResponse.OutputsEntry</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>key</td>
                  <td><a href="#string">string</a></td>
                  <td></td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>value</td>
                  <td><a href="#tensorflow.eas.ArrayProto">ArrayProto</a></td>
                  <td></td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="tensorflow.eas.ArrayDataType">ArrayDataType</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>DT_INVALID</td>
                <td>0</td>
                <td><pre>Not a legal value for DataType. Used to indicate a DataType field
has not been set.</pre></td>
              </tr>

              <tr>
                <td>DT_FLOAT</td>
                <td>1</td>
                <td><pre>Data types that all computation devices are expected to be
capable to support.</pre></td>
              </tr>

              <tr>
                <td>DT_DOUBLE</td>
                <td>2</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_INT32</td>
                <td>3</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_UINT8</td>
                <td>4</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_INT16</td>
                <td>5</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_INT8</td>
                <td>6</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_STRING</td>
                <td>7</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_COMPLEX64</td>
                <td>8</td>
                <td><pre>Single-precision complex</pre></td>
              </tr>

              <tr>
                <td>DT_INT64</td>
                <td>9</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_BOOL</td>
                <td>10</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_QINT8</td>
                <td>11</td>
                <td><pre>Quantized int8</pre></td>
              </tr>

              <tr>
                <td>DT_QUINT8</td>
                <td>12</td>
                <td><pre>Quantized uint8</pre></td>
              </tr>

              <tr>
                <td>DT_QINT32</td>
                <td>13</td>
                <td><pre>Quantized int32</pre></td>
              </tr>

              <tr>
                <td>DT_BFLOAT16</td>
                <td>14</td>
                <td><pre>Float32 truncated to 16 bits.  Only for cast ops.</pre></td>
              </tr>

              <tr>
                <td>DT_QINT16</td>
                <td>15</td>
                <td><pre>Quantized int16</pre></td>
              </tr>

              <tr>
                <td>DT_QUINT16</td>
                <td>16</td>
                <td><pre>Quantized uint16</pre></td>
              </tr>

              <tr>
                <td>DT_UINT16</td>
                <td>17</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_COMPLEX128</td>
                <td>18</td>
                <td><pre>Double-precision complex</pre></td>
              </tr>

              <tr>
                <td>DT_HALF</td>
                <td>19</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_RESOURCE</td>
                <td>20</td>
                <td><pre></pre></td>
              </tr>

              <tr>
                <td>DT_VARIANT</td>
                <td>21</td>
                <td><pre>Arbitrary C&#43;&#43; data types</pre></td>
              </tr>

          </tbody>
        </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/tower.proto">easy_rec/python/protos/tower.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.BayesTaskTower">BayesTaskTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>tower_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>task name for the task tower </pre></td>
                </tr>

                <tr>
                  <td>label_name</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>label for the task, default is label_fields by order </pre></td>
                </tr>

                <tr>
                  <td>metrics_set</td>
                  <td><a href="#protos.EvalMetrics">EvalMetrics</a></td>
                  <td>repeated</td>
                  <td><pre>metrics for the task </pre></td>
                </tr>

                <tr>
                  <td>loss_type</td>
                  <td><a href="#protos.LossType">LossType</a></td>
                  <td>optional</td>
                  <td><pre>loss for the task Default: CLASSIFICATION</pre></td>
                </tr>

                <tr>
                  <td>num_class</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>num_class for multi-class classification loss Default: 1</pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>task specific dnn </pre></td>
                </tr>

                <tr>
                  <td>relation_tower_names</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>related tower names </pre></td>
                </tr>

                <tr>
                  <td>relation_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>relation dnn </pre></td>
                </tr>

                <tr>
                  <td>weight</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>training loss weights Default: 1</pre></td>
                </tr>

                <tr>
                  <td>task_space_indicator_label</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>label name for indcating the sample space for the task tower </pre></td>
                </tr>

                <tr>
                  <td>in_task_space_weight</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>the loss weight for sample in the task space Default: 1</pre></td>
                </tr>

                <tr>
                  <td>out_task_space_weight</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>the loss weight for sample out the task space Default: 1</pre></td>
                </tr>

                <tr>
                  <td>losses</td>
                  <td><a href="#protos.Loss">Loss</a></td>
                  <td>repeated</td>
                  <td><pre>level for prediction
required uint32 prediction_level = 13;
prediction weights
optional float prediction_weight = 14 [default = 1.0];
multiple losses </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.TaskTower">TaskTower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>tower_name</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre>task name for the task tower </pre></td>
                </tr>

                <tr>
                  <td>label_name</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>label for the task, default is label_fields by order </pre></td>
                </tr>

                <tr>
                  <td>metrics_set</td>
                  <td><a href="#protos.EvalMetrics">EvalMetrics</a></td>
                  <td>repeated</td>
                  <td><pre>metrics for the task </pre></td>
                </tr>

                <tr>
                  <td>loss_type</td>
                  <td><a href="#protos.LossType">LossType</a></td>
                  <td>optional</td>
                  <td><pre>loss for the task Default: CLASSIFICATION</pre></td>
                </tr>

                <tr>
                  <td>num_class</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>num_class for multi-class classification loss Default: 1</pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>task specific dnn </pre></td>
                </tr>

                <tr>
                  <td>weight</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>training loss weights Default: 1</pre></td>
                </tr>

                <tr>
                  <td>task_space_indicator_label</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>label name for indcating the sample space for the task tower </pre></td>
                </tr>

                <tr>
                  <td>in_task_space_weight</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>the loss weight for sample in the task space Default: 1</pre></td>
                </tr>

                <tr>
                  <td>out_task_space_weight</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>the loss weight for sample out the task space Default: 1</pre></td>
                </tr>

                <tr>
                  <td>losses</td>
                  <td><a href="#protos.Loss">Loss</a></td>
                  <td>repeated</td>
                  <td><pre>multiple losses </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.Tower">Tower</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>input</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/train.proto">easy_rec/python/protos/train.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.IncrementSaveConfig">IncrementSaveConfig</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>sparse_save_secs</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>dense_save_secs</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>sparse_save_steps</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>dense_save_steps</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>debug_save_update</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>if open, will save increment updates to model_dir/incr_save/ Default: false</pre></td>
                </tr>

                <tr>
                  <td>kafka</td>
                  <td><a href="#protos.IncrementSaveConfig.Kafka">IncrementSaveConfig.Kafka</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>datahub</td>
                  <td><a href="#protos.IncrementSaveConfig.Datahub">IncrementSaveConfig.Datahub</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>fs</td>
                  <td><a href="#protos.IncrementSaveConfig.File">IncrementSaveConfig.File</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.IncrementSaveConfig.Datahub">IncrementSaveConfig.Datahub</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>akId</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>akSecret</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>region</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>project</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>topic</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>consumer</td>
                  <td><a href="#protos.IncrementSaveConfig.Datahub.Consumer">IncrementSaveConfig.Datahub.Consumer</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.IncrementSaveConfig.Datahub.Consumer">IncrementSaveConfig.Datahub.Consumer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>offset</td>
                  <td><a href="#int64">int64</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>timeout</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 600</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.IncrementSaveConfig.File">IncrementSaveConfig.File</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>incr_save_dir</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> Default: incr_save</pre></td>
                </tr>

                <tr>
                  <td>relative</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>relative to model_dir Default: true</pre></td>
                </tr>

                <tr>
                  <td>mount_path</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>for online inference, please set the storage.mount_path to mount_path
online service will fail Default: /home/admin/docker_ml/workspace/incr_save/</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.IncrementSaveConfig.Kafka">IncrementSaveConfig.Kafka</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>server</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>topic</td>
                  <td><a href="#string">string</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>consumer</td>
                  <td><a href="#protos.IncrementSaveConfig.Kafka.Consumer">IncrementSaveConfig.Kafka.Consumer</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.IncrementSaveConfig.Kafka.Consumer">IncrementSaveConfig.Kafka.Consumer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>config_topic</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>config_global</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>offset</td>
                  <td><a href="#int64">int64</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0</pre></td>
                </tr>

                <tr>
                  <td>timeout</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre> Default: 600</pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.TrainConfig">TrainConfig</h3>
        <pre>Message for configuring EasyRecModel training jobs (train.py).</pre><pre>Next id: 25</pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>optimizer_config</td>
                  <td><a href="#protos.Optimizer">Optimizer</a></td>
                  <td>repeated</td>
                  <td><pre>optimizer options </pre></td>
                </tr>

                <tr>
                  <td>gradient_clipping_by_norm</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>If greater than 0, clips gradients by this value. Default: 0</pre></td>
                </tr>

                <tr>
                  <td>num_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Number of steps to train the models: if 0, will train the model
indefinitely. Default: 0</pre></td>
                </tr>

                <tr>
                  <td>fine_tune_checkpoint</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>Checkpoint to restore variables from. </pre></td>
                </tr>

                <tr>
                  <td>fine_tune_ckpt_var_map</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>sync_replicas</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>Whether to synchronize replicas during training.
In case so, build a SyncReplicateOptimizer Default: true</pre></td>
                </tr>

                <tr>
                  <td>sparse_accumulator_type</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>only take effect on pai-tf when sync_replicas is set,
options are:
    raw, hash, multi_map, list, parallel
in general, multi_map runs faster than other options. Default: multi_map</pre></td>
                </tr>

                <tr>
                  <td>startup_delay_steps</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>Number of training steps between replica startup.
This flag must be set to 0 if sync_replicas is set to true. Default: 15</pre></td>
                </tr>

                <tr>
                  <td>save_checkpoints_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Step interval for saving checkpoint Default: 1000</pre></td>
                </tr>

                <tr>
                  <td>save_checkpoints_secs</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Seconds interval for saving checkpoint </pre></td>
                </tr>

                <tr>
                  <td>keep_checkpoint_max</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Max checkpoints to keep Default: 10</pre></td>
                </tr>

                <tr>
                  <td>save_summary_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>Save summaries every this many steps. Default: 1000</pre></td>
                </tr>

                <tr>
                  <td>log_step_count_steps</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>optional</td>
                  <td><pre>The frequency global step/sec and the loss will be logged during training. Default: 10</pre></td>
                </tr>

                <tr>
                  <td>is_profiling</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>profiling or not Default: false</pre></td>
                </tr>

                <tr>
                  <td>force_restore_shape_compatible</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>if variable shape is incompatible, clip or pad variables in checkpoint Default: false</pre></td>
                </tr>

                <tr>
                  <td>train_distribute</td>
                  <td><a href="#protos.DistributionStrategy">DistributionStrategy</a></td>
                  <td>optional</td>
                  <td><pre>DistributionStrategy, available values are &#39;mirrored&#39; and &#39;collective&#39; and &#39;ess&#39;
- mirrored: MirroredStrategy, single machine and multiple devices;
- collective: CollectiveAllReduceStrategy, multiple machines and multiple devices. Default: NoStrategy</pre></td>
                </tr>

                <tr>
                  <td>num_gpus_per_worker</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>Number of gpus per machine Default: 1</pre></td>
                </tr>

                <tr>
                  <td>summary_model_vars</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>summary model variables or not Default: false</pre></td>
                </tr>

                <tr>
                  <td>protocol</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>distribute training protocol [grpc&#43;&#43; | star_server]
grpc&#43;&#43;: https://help.aliyun.com/document_detail/173157.html?spm=5176.10695662.1996646101.searchclickresult.3ebf450evuaPT3
star_server: https://help.aliyun.com/document_detail/173154.html?spm=a2c4g.11186623.6.627.39ad7e3342KOX4 </pre></td>
                </tr>

                <tr>
                  <td>inter_op_parallelism_threads</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>inter_op_parallelism_threads Default: 0</pre></td>
                </tr>

                <tr>
                  <td>intra_op_parallelism_threads</td>
                  <td><a href="#int32">int32</a></td>
                  <td>optional</td>
                  <td><pre>intra_op_parallelism_threads Default: 0</pre></td>
                </tr>

                <tr>
                  <td>tensor_fuse</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>tensor fusion on PAI-TF Default: false</pre></td>
                </tr>

                <tr>
                  <td>write_graph</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>write graph into graph.pbtxt and summary or not Default: true</pre></td>
                </tr>

                <tr>
                  <td>freeze_gradient</td>
                  <td><a href="#string">string</a></td>
                  <td>repeated</td>
                  <td><pre>match variable patterns to freeze </pre></td>
                </tr>

                <tr>
                  <td>incr_save_config</td>
                  <td><a href="#protos.IncrementSaveConfig">IncrementSaveConfig</a></td>
                  <td>optional</td>
                  <td><pre>increment save config </pre></td>
                </tr>

                <tr>
                  <td>enable_oss_stop_signal</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>enable oss stop signal
stop by create OSS_STOP_SIGNAL under model_dir Default: false</pre></td>
                </tr>

                <tr>
                  <td>dead_line</td>
                  <td><a href="#string">string</a></td>
                  <td>optional</td>
                  <td><pre>stop training after dead_line time, format:
  20220508 23:59:59 </pre></td>
                </tr>

            </tbody>
          </table>


        <h3 id="protos.DistributionStrategy">DistributionStrategy</h3>
        <pre></pre>
        <table class="enum-table">
          <thead>
            <tr><td>Name</td><td>Number</td><td>Description</td></tr>
          </thead>
          <tbody>

              <tr>
                <td>NoStrategy</td>
                <td>0</td>
                <td><pre>use old SyncReplicasOptimizer for ParameterServer training</pre></td>
              </tr>

              <tr>
                <td>PSStrategy</td>
                <td>1</td>
                <td><pre>PSStrategy with multiple gpus on one node could not work
on pai-tf, could only work on TF &gt;=1.15</pre></td>
              </tr>

              <tr>
                <td>MirroredStrategy</td>
                <td>2</td>
                <td><pre>could only work on PaiTF or TF &gt;=1.15
single worker multiple gpu mode</pre></td>
              </tr>

              <tr>
                <td>CollectiveAllReduceStrategy</td>
                <td>3</td>
                <td><pre>Depreciated</pre></td>
              </tr>

              <tr>
                <td>ExascaleStrategy</td>
                <td>4</td>
                <td><pre>currently not working good</pre></td>
              </tr>

              <tr>
                <td>MultiWorkerMirroredStrategy</td>
                <td>5</td>
                <td><pre>multi worker multi gpu mode
see tf.distribute.experimental.MultiWorkerMirroredStrategy</pre></td>
              </tr>

          </tbody>
        </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/uniter.proto">easy_rec/python/protos/uniter.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.Uniter">Uniter</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>config</td>
                  <td><a href="#protos.UniterTower">UniterTower</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>final_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/variational_dropout.proto">easy_rec/python/protos/variational_dropout.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.VariationalDropoutLayer">VariationalDropoutLayer</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>regularization_lambda</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre>regularization coefficient lambda Default: 0.01</pre></td>
                </tr>

                <tr>
                  <td>embedding_wise_variational_dropout</td>
                  <td><a href="#bool">bool</a></td>
                  <td>optional</td>
                  <td><pre>variational_dropout dimension Default: false</pre></td>
                </tr>

            </tbody>
          </table>


      <div class="file-heading">
        <h2 id="easy_rec/python/protos/wide_and_deep.proto">easy_rec/python/protos/wide_and_deep.proto</h2><a href="#title">Top</a>
      </div>
      <pre></pre>


        <h3 id="protos.WideAndDeep">WideAndDeep</h3>
        <pre></pre>


          <table class="field-table">
            <thead>
              <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
            </thead>
            <tbody>

                <tr>
                  <td>wide_output_dim</td>
                  <td><a href="#uint32">uint32</a></td>
                  <td>required</td>
                  <td><pre> Default: 1</pre></td>
                </tr>

                <tr>
                  <td>dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>required</td>
                  <td><pre> </pre></td>
                </tr>

                <tr>
                  <td>final_dnn</td>
                  <td><a href="#protos.DNN">DNN</a></td>
                  <td>optional</td>
                  <td><pre>if set, the output of dnn and wide part are concatenated and
passed to the final_dnn; otherwise, they are summarized </pre></td>
                </tr>

                <tr>
                  <td>l2_regularization</td>
                  <td><a href="#float">float</a></td>
                  <td>optional</td>
                  <td><pre> Default: 0.0001</pre></td>
                </tr>

            </tbody>
          </table>


    <h2 id="scalar-value-types">Scalar Value Types</h2>
    <table class="scalar-value-types-table">
      <thead>
        <tr><td>.proto Type</td><td>Notes</td><td>C++ Type</td><td>Java Type</td><td>Python Type</td></tr>
      </thead>
      <tbody>

          <tr id="double">
            <td>double</td>
            <td></td>
            <td>double</td>
            <td>double</td>
            <td>float</td>
          </tr>

          <tr id="float">
            <td>float</td>
            <td></td>
            <td>float</td>
            <td>float</td>
            <td>float</td>
          </tr>

          <tr id="int32">
            <td>int32</td>
            <td>Uses variable-length encoding. Inefficient for encoding negative numbers – if your field is likely to have negative values, use sint32 instead.</td>
            <td>int32</td>
            <td>int</td>
            <td>int</td>
          </tr>

          <tr id="int64">
            <td>int64</td>
            <td>Uses variable-length encoding. Inefficient for encoding negative numbers – if your field is likely to have negative values, use sint64 instead.</td>
            <td>int64</td>
            <td>long</td>
            <td>int/long</td>
          </tr>

          <tr id="uint32">
            <td>uint32</td>
            <td>Uses variable-length encoding.</td>
            <td>uint32</td>
            <td>int</td>
            <td>int/long</td>
          </tr>

          <tr id="uint64">
            <td>uint64</td>
            <td>Uses variable-length encoding.</td>
            <td>uint64</td>
            <td>long</td>
            <td>int/long</td>
          </tr>

          <tr id="sint32">
            <td>sint32</td>
            <td>Uses variable-length encoding. Signed int value. These more efficiently encode negative numbers than regular int32s.</td>
            <td>int32</td>
            <td>int</td>
            <td>int</td>
          </tr>

          <tr id="sint64">
            <td>sint64</td>
            <td>Uses variable-length encoding. Signed int value. These more efficiently encode negative numbers than regular int64s.</td>
            <td>int64</td>
            <td>long</td>
            <td>int/long</td>
          </tr>

          <tr id="fixed32">
            <td>fixed32</td>
            <td>Always four bytes. More efficient than uint32 if values are often greater than 2^28.</td>
            <td>uint32</td>
            <td>int</td>
            <td>int</td>
          </tr>

          <tr id="fixed64">
            <td>fixed64</td>
            <td>Always eight bytes. More efficient than uint64 if values are often greater than 2^56.</td>
            <td>uint64</td>
            <td>long</td>
            <td>int/long</td>
          </tr>

          <tr id="sfixed32">
            <td>sfixed32</td>
            <td>Always four bytes.</td>
            <td>int32</td>
            <td>int</td>
            <td>int</td>
          </tr>

          <tr id="sfixed64">
            <td>sfixed64</td>
            <td>Always eight bytes.</td>
            <td>int64</td>
            <td>long</td>
            <td>int/long</td>
          </tr>

          <tr id="bool">
            <td>bool</td>
            <td></td>
            <td>bool</td>
            <td>boolean</td>
            <td>boolean</td>
          </tr>

          <tr id="string">
            <td>string</td>
            <td>A string must always contain UTF-8 encoded or 7-bit ASCII text.</td>
            <td>string</td>
            <td>String</td>
            <td>str/unicode</td>
          </tr>

          <tr id="bytes">
            <td>bytes</td>
            <td>May contain any arbitrary sequence of bytes.</td>
            <td>string</td>
            <td>ByteString</td>
            <td>str</td>
          </tr>

      </tbody>
    </table>
  </body>
</html>