ner_dstc2.json

{
  "dataset_reader": {
    "class_name": "dstc2_reader",
    "data_path": "{DATA_PATH}"
  },
  "dataset_iterator": {
    "class_name": "dstc2_ner_iterator",
    "slot_values_path": "{SLOT_VALS_PATH}"
  },
  "chainer": {
    "in": ["x"],
    "in_y": ["y"],
    "pipe": [
      {
        "in": ["x"],
        "class_name": "lazy_tokenizer",
        "out": ["x_tokens"]
      },
      {
        "in": ["x_tokens"],
        "class_name": "str_lower",
        "out": ["x_lower"]
      },
      {
        "in": ["x_lower"],
        "id": "word_vocab",
        "class_name": "simple_vocab",
        "pad_with_zeros": true,
        "fit_on": ["x_lower"],
        "save_path": "{MODEL_PATH}/word.dict",
        "load_path": "{MODEL_PATH}/word.dict",
        "out": ["x_tok_ind"]
      },
      {
        "class_name": "random_emb_mat",
        "id": "embeddings",
        "vocab_len": "#word_vocab.len",
        "emb_dim": 100
      },
      {
        "in": ["y"],
        "id": "tag_vocab",
        "class_name": "simple_vocab",
        "pad_with_zeros": true,
        "fit_on": ["y"],
        "save_path": "{MODEL_PATH}/tag.dict",
        "load_path": "{MODEL_PATH}/tag.dict",
        "out": ["y_ind"]
      },
      {
        "in": ["x_tokens"],
        "class_name": "mask",
        "out": ["mask"]
      },
      {
        "in": ["x_tok_ind", "mask"],
        "in_y": ["y_ind"],
        "out": ["y_predicted"],
        "class_name": "ner",
        "main": true,
        "token_emb_mat": "#embeddings.emb_mat",
        "n_hidden_list": [64, 64],
        "net_type": "cnn",
        "n_tags": "#tag_vocab.len",
        "save_path": "{MODEL_PATH}/model",
        "load_path": "{MODEL_PATH}/model",
        "embeddings_dropout": true,
        "top_dropout": true,
        "intra_layer_dropout": false,
        "use_batch_norm": true,
        "learning_rate": 1e-2,
        "dropout_keep_prob": 0.5
      },
      {
        "ref": "tag_vocab",
        "in": ["y_predicted"],
        "out": ["tags"]
      }
    ],
    "out": ["x_tokens", "tags"]
  },
  "train": {
    "epochs": 100,
    "batch_size": 64,

    "metrics": [
      {
        "name": "ner_f1",
        "inputs": ["y", "tags"]
      },
      {
        "name": "per_token_accuracy",
        "inputs": ["y", "tags"]
      }
    ],
    "validation_patience": 5,
    "val_every_n_epochs": 5,

    "log_every_n_batches": 100,
    "show_examples": false,
    "class_name": "nn_trainer",
    "evaluation_targets": [
      "valid",
      "test"
    ]
  },
  "metadata": {
    "variables": {
      "ROOT_PATH": "~/.deeppavlov",
      "DATA_PATH": "{ROOT_PATH}/downloads/dstc2",
      "SLOT_VALS_PATH": "{DATA_PATH}/dstc_slot_vals.json",
      "MODELS_PATH": "{ROOT_PATH}/models",
      "MODEL_PATH": "{MODELS_PATH}/slotfill_dstc2"
    },
    "requirements": [
      "{DEEPPAVLOV_PATH}/requirements/tf.txt"
    ],
    "download": [
      {
        "url": "http://files.deeppavlov.ai/deeppavlov_data/dstc_slot_vals.tar.gz",
        "subdir": "{DATA_PATH}"
      },
      {
        "url": "http://files.deeppavlov.ai/deeppavlov_data/slotfill_dstc2.tar.gz",
        "subdir": "{MODELS_PATH}"
      }
    ]
  }
}