Skip to content

Commit

Permalink
feat: multilingual and english BERT-based models for classification a…
Browse files Browse the repository at this point in the history 
…nd SQuAD (deeppavlov#742)

* feat: add multi_squad_retr dataset

* feat: add config file

* feat: add multi_squad_ru_retr dataset

* fix: paths and squad datareaders

* fix: squad preprocess

* feat: add context from squad rate sampling ratio

* fix: dataset reader

* fix: squad rate attr in squad_iterator

* feat: add link to trained noans_ru model

* feat: create odqa retr noans config, update top n for ru tfidf ranker

* fix: upd metrics docs

* feat: add bert for classification tasks

* fix: remove squad model from registry

* fix: add bert preprocessor

* fix: upd names in configs

* fix: set ru tfidf ranker top_n to 5

* fix: set paragraphs false in odqa

* fix: multilabel classification with bert

* fix: bert dropout config names

* feat: add AdamWeighDecay optimizer for bert

* config for rusentiment bert 1m ckpt

* chore: bert configs

* feat: add bert model for squad

* fix: squad bert config

* feat: add loading pretrained bert for squad

* feat: add multilingual bert for sbersquad

* feat: add url arg to squad dataset reader

* fix: bert answer postprocessor

* feat&fix: add bert squad infer model and config fixes

* chore: add squad bert infer config

* feat: add batching to bert inference model

* feat: add squad with rubert and odqa configs

* fix: squad rubert config

* chore: add ruodqa evaluation config

* fix: squad iterator empty batch

* chore: add one more bert based squad config

* feat: add noans score and model trained on retr dump

* chore: add inference config for bert on noans

* feat: add bert_ner (with bugs for now)

* fix: attribute assign in classifier

* fix: Adam inited twice

* fix: ner bert preprocessor

* refactor&docs: add docs for bert preprocessor

* feat: add bert_ner for inference

* docs: add bert preprocessor call method docs

* docs: bert squad model

* feat: add parametrization of taken encoder layers

* fix: rubert download link

* docs: train_on_batch and call for bert squad model

* docs: add docstring for BertSQuADInferModel

* fix: sigmoid -> softmax output

* docs: upd docstring

* fix: list -> tuple default value

* docs: add docstrings for bert classifier

* chore: rm configs

* feat: paraphraser config, model and scores

* feat: rusentiment with multi-lingual bert

* feat: rusentiment with multi-lingual bert

* feat: insults kaggle on English BERT

* chore: remove configs

* docs: add ru squad results

* feat: add bert models to tests

* docs: upd docs

* feat: upd configs and tests

* feat: rm bert configs

* refactor: rm debug logging

* fix: paraphraser dataset reader

* docs: update bert_ner docstring

* fix: restore tests for squad noans model

* fix: paraphraser tests

* fix: add bert_dp requirement to squad_bert config

* fix: download path for model files

* fix: rusentiment tests

* fix: import order

* fix: remove extra tabulation

* refactor: rename squad metrics

* refactor: url argument in squad_dataset_reader has higher priority than dataset

* refactor: remove process_event method from bert models

* feat: add bert-base model for EN SQuAD

* feat: add bert dstc2 model

* fix: squad bert infer config

* docs: add docs for NerPreprocessor

* feat: give higher priority to url arg in multi_squad_dataset_reader

* fix: bert NER preprocessor
  • Loading branch information
@yurakuratov @seliverstov
yurakuratov authored and seliverstov committed Mar 20, 2019
1 parent b57151e commit 174dec6
Show file tree
Hide file tree
Showing 35 changed files with 2,788 additions and 49 deletions.
144 changes: 144 additions & 0 deletions deeppavlov/configs/classifiers/insults_kaggle_bert.json
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,144 @@
{
"dataset_reader": {
"class_name": "basic_classification_reader",
"x": "Comment",
"y": "Class",
"data_path": "{DOWNLOADS_PATH}/insults_data"
},
"dataset_iterator": {
"class_name": "basic_classification_iterator",
"seed": 42
},
"chainer": {
"in": [
"x"
],
"in_y": [
"y"
],
"pipe": [
{
"class_name": "bert_preprocessor",
"vocab_file": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/vocab.txt",
"do_lower_case": false,
"max_seq_length": 64,
"in": [
"x"
],
"out": [
"bert_features"
]
},
{
"id": "classes_vocab",
"class_name": "simple_vocab",
"fit_on": [
"y"
],
"save_path": "{MODELS_PATH}/classes.dict",
"load_path": "{MODELS_PATH}/classes.dict",
"in": "y",
"out": "y_ids"
},
{
"in": "y_ids",
"out": "y_onehot",
"class_name": "one_hotter",
"depth": "#classes_vocab.len",
"single_vector": true
},
{
"class_name": "bert_classifier",
"n_classes": "#classes_vocab.len",
"return_probas": true,
"one_hot_labels": true,
"bert_config_file": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/bert_config.json",
"pretrained_bert": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/bert_model.ckpt",
"save_path": "{MODELS_PATH}/model",
"load_path": "{MODELS_PATH}/model",
"keep_prob": 0.5,
"learning_rate": 1e-05,
"learning_rate_drop_patience": 5,
"learning_rate_drop_div": 2.0,
"in": [
"bert_features"
],
"in_y": [
"y_onehot"
],
"out": [
"y_pred_probas"
]
},
{
"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": 100,
"batch_size": 64,
"metrics": [
{
"name": "roc_auc",
"inputs": [
"y_onehot",
"y_pred_probas"
]
},
"sets_accuracy",
"f1_macro"
],
"validation_patience": 5,
"val_every_n_epochs": 1,
"log_every_n_epochs": 1,
"show_examples": false,
"evaluation_targets": [
"train",
"valid",
"test"
],
"class_name": "nn_trainer",
"tensorboard_log_dir": "{MODELS_PATH}/"
},
"metadata": {
"variables": {
"ROOT_PATH": "~/.deeppavlov",
"DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
"MODELS_PATH": "{ROOT_PATH}/models/classifiers/insults_kaggle_v3"
},
"requirements": [
"{DEEPPAVLOV_PATH}/requirements/tf.txt",
"{DEEPPAVLOV_PATH}/requirements/bert_dp.txt"
],
"labels": {
"telegram_utils": "IntentModel",
"server_utils": "KerasIntentModel"
},
"download": [
{
"url": "http://files.deeppavlov.ai/datasets/insults_data.tar.gz",
"subdir": "{DOWNLOADS_PATH}"
},
{
"url": "http://files.deeppavlov.ai/deeppavlov_data/bert/cased_L-12_H-768_A-12.zip",
"subdir": "{DOWNLOADS_PATH}/bert_models"
},
{
"url": "http://files.deeppavlov.ai/deeppavlov_data/classifiers/insults_kaggle_v3.tar.gz",
"subdir": "{ROOT_PATH}/models/classifiers"
}
]
}
}
128 changes: 128 additions & 0 deletions deeppavlov/configs/classifiers/intents_dstc2_bert.json
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,128 @@
{
"dataset_reader": {
"class_name": "dstc2_reader",
"data_path": "{DOWNLOADS_PATH}/dstc2"
},
"dataset_iterator": {
"class_name": "dstc2_intents_iterator",
"seed": 42
},
"chainer": {
"in": ["x"],
"in_y": ["y"],
"pipe": [
{
"id": "classes_vocab",
"class_name": "simple_vocab",
"fit_on": ["y"],
"save_path": "{MODELS_PATH}/classes.dict",
"load_path": "{MODELS_PATH}/classes.dict",
"in": "y",
"out": "y_ids",
"special_tokens": ["<UNK>"]
},
{
"class_name": "bert_preprocessor",
"vocab_file": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/vocab.txt",
"do_lower_case": false,
"max_seq_length": 64,
"in": ["x"],
"out": ["bert_features"]
},
{
"in": "y_ids",
"out": "y_onehot",
"class_name": "one_hotter",
"id": "my_one_hotter",
"depth": "#classes_vocab.len",
"single_vector": true
},
{
"class_name": "bert_classifier",
"n_classes": "#classes_vocab.len",
"return_probas": true,
"one_hot_labels": true,
"multilabel": true,
"bert_config_file": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/bert_config.json",
"pretrained_bert": "{DOWNLOADS_PATH}/bert_models/cased_L-12_H-768_A-12/bert_model.ckpt",
"save_path": "{MODELS_PATH}/model",
"load_path": "{MODELS_PATH}/model",
"keep_prob": 0.5,
"learning_rate": 2e-05,
"learning_rate_drop_patience": 3,
"learning_rate_drop_div": 2.0,
"in": ["bert_features"],
"in_y": ["y_onehot"],
"out": ["y_pred_probas"]
},
{
"in": "y_pred_probas",
"out": "y_pred_ids",
"class_name": "proba2labels",
"confident_threshold": 0.5
},
{
"in": "y_pred_ids",
"out": "y_pred_labels",
"ref": "classes_vocab"
},
{
"ref": "my_one_hotter",
"in": "y_pred_ids",
"out": "y_pred_onehot"
}
],
"out": ["y_pred_probas", "y_pred_labels"]
},
"train": {
"metrics": [
{
"name": "sets_accuracy",
"inputs": ["y", "y_pred_labels"]
},
{
"name": "roc_auc",
"inputs": ["y_onehot", "y_pred_probas"]
}
],
"show_examples": false,
"batch_size": 32,
"pytest_max_batches": 2,
"validation_patience": 10,
"val_every_n_batches": 100,
"log_every_n_batches": 100,
"validate_best": true,
"test_best": true,
"tensorboard_log_dir": "{MODELS_PATH}/logs"
},
"metadata": {
"variables": {
"ROOT_PATH": "~/.deeppavlov",
"DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
"MODELS_PATH": "{ROOT_PATH}/models/classifiers/intents_dstc2_bert_v0"
},
"requirements": [
"{DEEPPAVLOV_PATH}/requirements/tf.txt",
"{DEEPPAVLOV_PATH}/requirements/bert_dp.txt"
],
"labels": {
"telegram_utils": "IntentModel",
"server_utils": "KerasIntentModel"
},
"download": [
{
"url": "http://files.deeppavlov.ai/deeppavlov_data/bert/cased_L-12_H-768_A-12.zip",
"subdir": "{DOWNLOADS_PATH}/bert_models"
},
{
"url": "http://files.deeppavlov.ai/datasets/dstc2_v2.tar.gz",
"subdir": "{DOWNLOADS_PATH}/dstc2"
},
{
"url": "http://files.deeppavlov.ai/deeppavlov_data/classifiers/intents_dstc2_bert_v0.tar.gz",
"subdir": "{ROOT_PATH}/models/classifiers"
}

]
}
}
107 changes: 107 additions & 0 deletions deeppavlov/configs/classifiers/paraphraser_bert.json
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,107 @@
{
"dataset_reader": {
"class_name": "paraphraser_reader",
"data_path": "{DOWNLOADS_PATH}/paraphraser_data",
"do_lower_case": false
},
"dataset_iterator": {
"class_name": "siamese_iterator",
"seed": 243,
"len_valid": 500
},
"chainer": {
"in": [
"text_a",
"text_b"
],
"in_y": [
"y"
],
"pipe": [
{
"class_name": "bert_preprocessor",
"vocab_file": "{DOWNLOADS_PATH}/bert_models/multi_cased_L-12_H-768_A-12/vocab.txt",
"do_lower_case": false,
"max_seq_length": 64,
"in": [
"text_a",
"text_b"
],
"out": [
"bert_features"
]
},
{
"class_name": "bert_classifier",
"n_classes": 2,
"one_hot_labels": false,
"bert_config_file": "{DOWNLOADS_PATH}/bert_models/multi_cased_L-12_H-768_A-12/bert_config.json",
"pretrained_bert": "{DOWNLOADS_PATH}/bert_models/multi_cased_L-12_H-768_A-12/bert_model.ckpt",
"save_path": "{MODELS_PATH}/model_multi",
"load_path": "{MODELS_PATH}/model_multi",
"keep_prob": 0.5,
"learning_rate": 2e-05,
"learning_rate_drop_patience": 2,
"learning_rate_drop_div": 2.0,
"in": [
"bert_features"
],
"in_y": [
"y"
],
"out": [
"predictions"
]
}
],
"out": [
"predictions"
]
},
"train": {
"batch_size": 32,
"pytest_max_batches": 2,
"metrics": [
"f1",
"acc"
],
"validation_patience": 10,
"val_every_n_batches": 100,
"log_every_n_batches": 100,
"evaluation_targets": [
"train",
"valid",
"test"
],
"tensorboard_log_dir": "{MODELS_PATH}/"
},
"metadata": {
"variables": {
"ROOT_PATH": "~/.deeppavlov",
"DOWNLOADS_PATH": "{ROOT_PATH}/downloads",
"MODELS_PATH": "{ROOT_PATH}/models/paraphraser_bert_v0"
},
"requirements": [
"{DEEPPAVLOV_PATH}/requirements/tf.txt",
"{DEEPPAVLOV_PATH}/requirements/bert_dp.txt"
],
"download": [
{
"url": "http://files.deeppavlov.ai/datasets/paraphraser.zip",
"subdir": "{DOWNLOADS_PATH}/paraphraser_data"
},
{
"url": "http://files.deeppavlov.ai/datasets/paraphraser_gold.zip",
"subdir": "{DOWNLOADS_PATH}/paraphraser_data"
},
{
"url": "http://files.deeppavlov.ai/deeppavlov_data/bert/multi_cased_L-12_H-768_A-12.zip",
"subdir": "{DOWNLOADS_PATH}/bert_models"
},
{
"url": "http://files.deeppavlov.ai/deeppavlov_data/classifiers/paraphraser_bert_v0.tar.gz",
"subdir": "{ROOT_PATH}/models"
}
]
}
}
Loading

0 comments on commit 174dec6

Please sign in to comment.