ranking_ubuntu_v1_mt_word2vec_dam_transformer.json

{
  "info": "The config is for training or evaluation of DAM_USE-T on Ubuntu Dialogue Corpus v1 using prepared Word2vec embeddings",
  "dataset_reader": {
    "class_name": "ubuntu_v1_mt_reader",
    "data_path": "{DOWNLOADS_PATH}/ubuntu_v1_data",
    "num_context_turns": "{NUM_CONTEXT_TURNS}",
    "padding": "pre"
  },
  "dataset_iterator": {
    "class_name": "siamese_iterator",
    "shuffle": true,
    "seed": 42
  },
  "chainer": {
    "in": ["x"],
    "in_y": ["y"],
    "pipe": [
      {
        "class_name": "split_tokenizer",
        "id": "tok_1"
      },
      {
        "class_name": "simple_vocab",
        "special_tokens": ["", "<UNK>"],
        "unk_token": "<UNK>",
        "fit_on": ["x"],
        "id": "vocab_1",
        "save_path": "{MODELS_PATH}/ubuntu_v1_mt_word2vec_dam_transformer/vocabs/int_tok.dict",
        "load_path": "{MODELS_PATH}/ubuntu_v1_mt_word2vec_dam_transformer/vocabs/int_tok.dict"
      },
      {
        "id": "word2vec_embedder",
        "class_name": "glove",
        "dim": 200,
        "load_path": "{DOWNLOADS_PATH}/embeddings/v1_ubuntu_word2vec_200_dam_0.779.txt"
      },
      {
        "id": "preproc",
        "class_name": "siamese_preprocessor",
        "save_path": "{MODELS_PATH}/ubuntu_v1_mt_word2vec_dam_transformer/preproc/tok.dict",
        "load_path": "{MODELS_PATH}/ubuntu_v1_mt_word2vec_dam_transformer/preproc/tok.dict",
        "num_ranking_samples": 10,
        "num_context_turns": "{NUM_CONTEXT_TURNS}",
        "max_sequence_length": 50,
        "embedding_dim": 200,
        "add_raw_text": true,
        "fit_on": ["x"],
        "in": ["x"],
        "out": ["x_proc"],
        "tokenizer": {
          "ref": "tok_1",
          "notes": "use defined tokenizer"
        },
        "vocab": {
          "ref": "vocab_1",
          "notes": "use vocab built for tokenized data"
        }
      },
      {
        "id": "embeddings",
        "class_name": "emb_mat_assembler",
        "embedder": "#word2vec_embedder",
        "vocab": "#vocab_1"
      },
      {
        "in": ["x_proc"],
        "in_y": ["y"],
        "out": ["y_predicted"],
        "class_name": "dam_nn_use_transformer",
        "stack_num": 5,
        "is_positional": true,
        "num_context_turns": "{NUM_CONTEXT_TURNS}",
        "max_sequence_length": "#preproc.max_sequence_length",
        "embedding_dim": "#word2vec_embedder.dim",
        "emb_matrix": "#embeddings.emb_mat",
        "learning_rate": 1e-3,
        "batch_size": 100,
        "seed": 42,
        "decay_steps": 2000,
        "save_path": "{MODELS_PATH}/ubuntu_v1_mt_word2vec_dam_transformer/model_dam/model",
        "load_path": "{MODELS_PATH}/ubuntu_v1_mt_word2vec_dam_transformer/model_dam/model"
      }
    ],
    "out": [
      "y_predicted"
    ]
  },
  "train": {
    "class_name": "nn_trainer",
    "epochs": 8,
    "batch_size": 100,
    "shuffle": true,
    "pytest_max_batches": 2,
    "train_metrics": [],
    "validate_best": true,
    "test_best": true,
    "metrics": [
      "r@1",
      "r@2",
      "r@5",
      "rank_response"
    ],
    "validation_patience": 1,
    "val_every_n_epochs": 1,
    "log_every_n_batches": 100,
    "evaluation_targets": [
      "valid",
      "test"
    ],
    "tensorboard_log_dir": "{MODELS_PATH}/ubuntu_v1_mt_word2vec_dam_transformer/logs_dam/"
  },
  "metadata": {
    "variables": {
      "ROOT_PATH": "~/.deeppavlov",
      "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
      "MODELS_PATH": "{ROOT_PATH}/models",
      "NUM_CONTEXT_TURNS": 10
    },
    "requirements": [
      "{DEEPPAVLOV_PATH}/requirements/tf.txt",
      "{DEEPPAVLOV_PATH}/requirements/tf-hub.txt",
      "{DEEPPAVLOV_PATH}/requirements/gensim.txt"
    ],
    "download": [
      {
        "url": "http://files.deeppavlov.ai/deeppavlov_data/ubuntu_v1_mt_word2vec_dam_transformer.tar.gz",
        "subdir": "{MODELS_PATH}"
      },
      {
        "url": "http://files.deeppavlov.ai/datasets/ubuntu_v1_data.tar.gz",
        "subdir": "{DOWNLOADS_PATH}/ubuntu_v1_data"
      },
      {
        "url": "http://files.deeppavlov.ai/embeddings/v1_ubuntu_word2vec_200_dam_0.779.txt.tar.gz",
        "subdir": "{DOWNLOADS_PATH}/embeddings"
      }
    ]
  }
}