custom_config.json

{
  "dataset_reader": {
    "class_name": "basic_classification_reader",
    "x": "text",
    "y": "sentiment",
    "data_path": "/train/",
    "train": "train.csv",
    "valid": "valid.csv"
  },
  "dataset_iterator": {
    "class_name": "basic_classification_iterator",
    "seed": 42
  },
  "chainer": {
    "in": [
      "x"
    ],
    "in_y": [
      "y"
    ],
    "pipe": [
      {
        "id": "classes_vocab",
        "class_name": "simple_vocab",
        "fit_on": [
          "y"
        ],
        "save_path": "custom_bert/classes.dict",
        "load_path": "custom_bert/classes.dict",
        "in": "y",
        "out": "y_ids"
      },
      {
        "class_name": "transformers_bert_preprocessor",
        "vocab_file": "vocab.txt",
        "do_lower_case": false,
        "max_seq_length": 512,
        "in": ["x"],
        "out": ["tokens", "subword_tokens", "subword_tok_ids", "startofword_markers", "attention_mask"]
      },
      {
        "class_name": "transformers_bert_embedder",
        "id": "my_embedder",
        "bert_config_path": "config.json",
        "truncate": false,
        "load_path": "custom_bert",
        "in": ["subword_tok_ids", "startofword_markers", "attention_mask"],
        "out": ["word_emb", "subword_emb", "max_emb", "mean_emb", "pooler_output"]
      },
      {
        "in": "y_ids",
        "out": "y_onehot",
        "class_name": "one_hotter",
        "depth": "#classes_vocab.len",
        "single_vector": true
      },
      {
        "in": [
          "word_emb"
        ],
        "in_y": [
          "y_onehot"
        ],
        "out": [
          "y_pred_probas"
        ],
        "main": true,
        "class_name": "keras_classification_model",
        "save_path": "custom_bert/model",
        "load_path": "custom_bert/model",
        "embedding_size": "#my_embedder.dim",
        "n_classes": "#classes_vocab.len",
        "kernel_sizes_cnn": [
          3,
          5,
          7
        ],
        "filters_cnn": 512,
        "optimizer": "Adam",
        "learning_rate": 0.01,
        "learning_rate_decay": 0.1,
        "loss": "categorical_crossentropy",
        "last_layer_activation": "softmax",
        "coef_reg_cnn": 1e-3,
        "coef_reg_den": 1e-2,
        "dropout_rate": 0.5,
        "dense_size": 100,
        "model_name": "cnn_model"
      },
      {
        "in": "y_pred_probas",
        "out": "y_pred_ids",
        "class_name": "proba2labels",
        "max_proba": true
      },
      {
        "in": "y_pred_ids",
        "out": "y_pred_labels",
        "ref": "classes_vocab"
      }
    ],
    "out": [
      "y_pred_labels"
    ]
  },
  "train": {
    "epochs": 3,
    "batch_size": 16,
    "metrics": [
      "accuracy",
      "f1_macro",
      "f1_weighted",
      {
        "name": "roc_auc",
        "inputs": ["y_onehot", "y_pred_probas"]
      }
    ],
    "validation_patience": 1,
    "val_every_n_epochs": 1,
    "log_every_n_epochs": 1,
    "show_examples": false,
    "evaluation_targets": [
      "train",
      "valid"
    ],
    "class_name": "nn_trainer"
  }
}