v0.3.1

Major Features and Improvements

• Refactor:
  • Layers in deepctr.layers are moved to deepctr.layers.core,deepctr.layers.interaction and deepctr.layers.normalization
  • Layers in deepctr.activations are moved to deepctr.layers.activation
  • Layers in deepctr.sequence are moved to deepctr.layers.sequence
• Add BiLSTM and Transformer in deepctr.layers.sequence docs

API changes

• deepctr.layers, deepctr.activations, deepctr.sequence change as shown above.

v0.2.3

Major Features and Improvements

• Adopted a new feature config generation method. example
• Simplified input mechanism for multi-valued features.example

Bug Fixes and Other Changes

• In some cases the order of model input placeholder and actual model input list does not match.
• Divide by zero overflow in SequencePoolingLayer when sequence length is 0.

API changes

• Now we must use SingleFeat and VarLenFeat to build feature configuration dict.Previous versions of the code require a little modification to work properly. example

v0.2.2

Major Features and Improvements

• Add sequence(multi-value) input support for AFM,AutoInt,DCN,DeepFM,FNN,NFM,PNN,xDeepFM models
• Refactor Input&Embedding module

v0.2.1

Major Features and Improvements

• Add AutoInt 中文模型介绍
  ) model.
• Add deepctr.layers.InteractingLayer

v0.2.0

Major Features and Improvements

• Add xDeepFM model.
• Add automatic check for new version of pypi package

Bug Fixes and Other Changes

• Fix some bugs when model's input field size is 1.
• Encapsulate some duplicate code

v0.1.6

Major Features and Improvements

• deepctr now is compatible with tensorflow from 1.4.0-1.12.0 except for 1.7.*and1.8.*.
• Add normalized_att_score attribute in deepctr.layers.AFMLayer,so we can get the attentional weight distribution of feature interactions in deepctr.models.AFM.
• Refactor deepctr.models.MLR
• Update deepctr.models.PredictionLayer to support Layer type activations.

Bug Fixes and Other Changes

• Fix bug in deepctr.layers.DCN when only use CrossNet.
• Fix bug in get_config of deepctr.layers.LocalActivationUnit

API changes

• Add default args to deepctr.layers.MLP and deepctr.layers.LocalActivationUnit

v0.1.5

Major Features and Improvements

• Modify output shape of BiInteractionPooling, InnerProductLayer to make
  sure that the dimensions of the output and input of a layer are same if possible

Bug Fixes and Other Changes

• Minimize the nesting of other layers in the custom layer, because I found that the statistics of the parameters number of model.summary() are incorrect when using other layers.

v0.1.4

Major Features and Improvements

• Use a more space efficient implementation in deepctr.layers.CrossNet
• Add linear and bias term in deepctr.models.FNN
• Add custom_objects in deepctr.utils which will be used in tf.keras.models.load_model

Bug Fixes and Other Changes

• Fix bug in deepctr.layers.MLP when use advance activation layer