dream

1
{
2
    "dataset_reader": {
3
      "class_name": "intents_dataset_reader:IntentsJsonReader",
4
      "data_path": "./",
5
      "train": "intent_phrases.json",
6
      "generated_data_path": "./generated_data"
7
    },
8
    "dataset_iterator": {
9
        "class_name": "basic_classification_iterator",
10
        "seed": 42,
11
        "split_seed": 23,
12
        "field_to_split": "train",
13
        "split_fields": [
14
          "train",
15
          "valid"
16
        ],
17
        "split_proportions": [
18
          0.8,
19
          0.2
20
        ]
21
    },
22
    "chainer": {
23
      "in": [
24
        "x"
25
      ],
26
      "in_y": [
27
        "y"
28
      ],
29
      "pipe": [
30
        {
31
          "class_name": "torch_transformers_preprocessor",
32
          "vocab_file": "{TRANSFORMER}",
33
          "do_lower_case": true,
34
          "max_seq_length": 64,
35
          "in": [
36
            "x"
37
          ],
38
          "out": [
39
            "bert_features"
40
          ]
41
        },
42
        {
43
          "id": "classes_vocab",
44
          "class_name": "simple_vocab",
45
          "fit_on": [
46
            "y"
47
          ],
48
          "save_path": "{MODEL_PATH}/classes.dict",
49
          "load_path": "{MODEL_PATH}/classes.dict",
50
          "in": [
51
            "y"
52
          ],
53
          "out": [
54
            "y_ids"
55
          ]
56
        },
57
        {
58
          "id": "my_one_hotter",
59
          "in": [
60
            "y_ids"
61
          ],
62
          "out": [
63
            "y_onehot"
64
          ],
65
          "class_name": "one_hotter",
66
          "depth": "#classes_vocab.len",
67
          "single_vector": true
68
        },
69
        {
70
          "class_name": "torch_transformers_classifier",
71
          "n_classes": "#classes_vocab.len",
72
          "return_probas": true,
73
          "one_hot_labels": true,
74
          "multilabel": true,
75
          "pretrained_bert": "{TRANSFORMER}",
76
          "save_path": "{MODEL_PATH}/model",
77
          "load_path": "{MODEL_PATH}/model",
78
          "optimizer": "AdamW",
79
          "optimizer_parameters": {
80
            "lr": 1e-05
81
          },
82
          "learning_rate_drop_patience": 5,
83
          "learning_rate_drop_div": 2.0,
84
          "in": [
85
            "bert_features"
86
          ],
87
          "in_y": [
88
            "y_onehot"
89
          ],
90
          "out": [
91
            "y_pred_probas"
92
          ]
93
        },
94
        {
95
          "in": [
96
            "y_pred_probas"
97
          ],
98
          "out": [
99
            "y_pred_ids"
100
          ],
101
          "class_name": "proba2labels",
102
          "max_proba": false,
103
          "confidence_threshold": 0.5
104
        },        
105
        {
106
          "ref": "my_one_hotter",
107
          "in": [
108
            "y_pred_ids"
109
          ],
110
          "out": [
111
            "y_pred_onehot"
112
          ]
113
        },
114
        {
115
          "in": [
116
            "y_pred_ids"
117
          ],
118
          "out": [
119
            "y_pred_labels"
120
          ],
121
          "ref": "classes_vocab"
122
        }
123
      ],
124
      "out": [
125
        "y_pred_labels",
126
        "y_pred_probas"
127
      ]
128
    },
129
    "train": {
130
      "epochs": 5,
131
      "batch_size": 64,
132
      "metrics": [
133
        {
134
          "name": "accuracy",
135
          "inputs": [
136
            "y",
137
            "y_pred_labels"
138
          ]
139
        },
140
        {
141
          "name": "f1_weighted",
142
          "inputs": [
143
            "y_onehot",
144
            "y_pred_onehot"
145
          ]
146
        },
147
        {
148
            "name": "f1_macro",
149
            "inputs": [
150
              "y_onehot",
151
              "y_pred_onehot"
152
            ]
153
        },
154
        {
155
          "name": "roc_auc",
156
          "inputs": [
157
            "y_onehot",
158
            "y_pred_probas"
159
          ]
160
        }
161
      ],
162
      "validation_patience": 5,
163
      "val_every_n_epochs": 1,
164
      "log_every_n_epochs": 1,
165
      "show_examples": false,
166
      "evaluation_targets": [
167
        "train",
168
        "valid"
169
      ],
170
      "class_name": "torch_trainer"
171
    },
172
    "metadata": {
173
      "imports": [
174
        "intents_dataset_reader"
175
      ],
176
      "variables": {
177
        "TRANSFORMER": "distilbert-base-uncased",
178
        "ROOT_PATH": "~/.deeppavlov",
179
        "DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
180
        "MODELS_PATH": "{ROOT_PATH}/models",
181
        "MODEL_PATH": "{MODELS_PATH}/classifiers/intents_model_v2"
182
      },
183
      "download": [
184
        {
185
          "url": "http://files.deeppavlov.ai/deeppavlov_data/intents_model_v2.tar.gz",
186
          "subdir": "{MODELS_PATH}/classifiers"
187
        }
188
      ]
189
    }
190
  }
191