model/train_joint_bert.py at main · HumaniCare/model · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
# -*- coding: utf-8 -*-
"""
@author: demiust
"""

import argparse
import os
import pickle

import numpy as np
import tensorflow as tf

from sklearn.preprocessing import LabelEncoder

from models.joint_bert import JointBertModel
from readers.goo_format_reader import Reader
from vectorizers.bert_vectorizer import BERTVectorizer
from vectorizers.tags_vectorizer import TagsVectorizer

# read command-line parameters
parser = argparse.ArgumentParser('Training the Joint BERT NLU model')
parser.add_argument('--train', '-t', help = 'Path to training data in Goo et al format', type = str, required = True)
parser.add_argument('--val', '-v', help = 'Path to validation data in Goo et al format', type = str, default = "", required = False)
parser.add_argument('--save', '-s', help = 'Folder path to save the trained model', type = str, required = True)
parser.add_argument('--epochs', '-e', help = 'Number of epochs', type = int, default = 5, required = False)
parser.add_argument('--batch', '-bs', help = 'Batch size', type = int, default = 64, required = False)
#parser.add_argument('--type', '-tp', help = 'bert   or    albert', type = str, default = 'bert', required = False)


#VALID_TYPES = ['bert', 'albert']

args = parser.parse_args()
train_data_folder_path = args.train
val_data_folder_path = args.val
save_folder_path = args.save
epochs = args.epochs
batch_size = args.batch
#type_ = args.type
print('train_data_folder_path:', train_data_folder_path)

## this line is to disable gpu
#os.environ["CUDA_VISIBLE_DEVICES"]="-1"


tf.compat.v1.random.set_random_seed(7)

config = tf.ConfigProto(intra_op_parallelism_threads=0,
                        inter_op_parallelism_threads=0,
                        allow_soft_placement=True,
                        device_count = {'GPU': 1})
sess = tf.compat.v1.Session(config=config)

#if type_ == 'bert':
#    bert_model_hub_path = "./bert-module" # to use KorBert by Etri
#    is_bert = True
#elif type_ == 'albert':
#    bert_model_hub_path = './korwiki_mecab_module'
#    is_bert = False
#else:
#    raise ValueError('type must be one of these values: %s' % str(VALID_TYPES))

bert_model_hub_path = './albert-module'
is_bert = False

print('read data ...')


# validation path가 잘 설정되어 있으면 val도 같이 train
if val_data_folder_path:
    train_text_arr, train_tags_arr, train_intents = Reader.read(train_data_folder_path)
    val_text_arr, val_tags_arr, val_intents = Reader.read(val_data_folder_path)

    print('train_text_arr[0:2] :', train_text_arr[0:2])
    print('train_tags_arr[0:2] :', train_tags_arr[0:2])
    print('train_intents[0:2] :', train_intents[0:2])

    print('vectorize data ...')
    bert_vectorizer = BERTVectorizer(sess, is_bert, bert_model_hub_path)
    # now bert model hub path exists --> already tokenized dataset
    # bert vectorizer MUST NOT tokenize input !!!

    print('bert vectorizer started ...') #valid pos removed
    train_input_ids, train_input_mask, train_segment_ids, train_sequence_lengths = bert_vectorizer.transform(train_text_arr)
    val_input_ids, val_input_mask, val_segment_ids, val_sequence_lengths = bert_vectorizer.transform(val_text_arr)

    print('vectorize tags ...')
    tags_vectorizer = TagsVectorizer()
    tags_vectorizer.fit(train_tags_arr)
    train_tags = tags_vectorizer.transform(train_tags_arr, train_input_ids)
    #val_tags = tags_vectorizer.transform(val_tags_arr, val_valid_positions)
    val_tags = tags_vectorizer.transform(val_tags_arr, val_input_ids)
    print('train_tags :', train_tags[0:2])
    slots_num = len(tags_vectorizer.label_encoder.classes_)
    print('slot num :', slots_num, tags_vectorizer.label_encoder.classes_)
    print('encode labels ...')
    intents_label_encoder = LabelEncoder()
    train_intents = intents_label_encoder.fit_transform(train_intents).astype(np.int32)
    #val_intents = intents_label_encoder.transform(val_intents).astype(np.int32)
    val_intents = intents_label_encoder.transform(val_intents).astype(np.int32)
    intents_num = len(intents_label_encoder.classes_)
    print('intents num :', intents_num)

    model = JointBertModel(slots_num, intents_num, bert_model_hub_path, sess,
                        num_bert_fine_tune_layers=10, is_bert=is_bert)


    print('train input shape :', train_input_ids.shape, train_input_ids[0:2])
    print('train_input_mask :', train_input_mask.shape, train_input_mask[0:2])
    print('train_segment_ids :', train_segment_ids.shape, train_segment_ids[0:2])
    print('train_tags :', train_tags.shape, train_tags[0:2])
    print('train_intents :', train_intents.shape, train_intents[0:2])


    print('training model ...')

    model.fit([train_input_ids, train_input_mask, train_segment_ids], [train_tags, train_intents],
            validation_data=([val_input_ids, val_input_mask, val_segment_ids], [val_tags, val_intents]),
            epochs=epochs, batch_size=batch_size)

# validation 없을 경우 train데이터로만!
else:
    train_text_arr, train_tags_arr, train_intents = Reader.read(train_data_folder_path)
    #val_text_arr, val_tags_arr, val_intents = Reader.read(val_data_folder_path)

    print('train_text_arr[0:2] :', train_text_arr[0:2])
    print('train_tags_arr[0:2] :', train_tags_arr[0:2])
    print('train_intents[0:2] :', train_intents[0:2])

    print('vectorize data ...')
    bert_vectorizer = BERTVectorizer(sess, is_bert, bert_model_hub_path)
    # now bert model hub path exists --> already tokenized dataset
    # bert vectorizer MUST NOT tokenize input !!!

    print('bert vectorizer started ...') #valid pos removed
    train_input_ids, train_input_mask, train_segment_ids, train_sequence_lengths = bert_vectorizer.transform(train_text_arr)
    #val_input_ids, val_input_mask, val_segment_ids, val_valid_positions, val_sequence_lengths = bert_vectorizer.transform(val_text_arr)

    print('vectorize tags ...')
    tags_vectorizer = TagsVectorizer()
    tags_vectorizer.fit(train_tags_arr)
    train_tags = tags_vectorizer.transform(train_tags_arr, train_input_ids)
    #val_tags = tags_vectorizer.transform(val_tags_arr, val_valid_positions)
    print('train_tags :', train_tags[0:2])
    slots_num = len(tags_vectorizer.label_encoder.classes_)
    print('slot num :', slots_num, tags_vectorizer.label_encoder.classes_)

    print('encode labels ...')
    intents_label_encoder = LabelEncoder()
    train_intents = intents_label_encoder.fit_transform(train_intents).astype(np.int32)
    #val_intents = intents_label_encoder.transform(val_intents).astype(np.int32)
    intents_num = len(intents_label_encoder.classes_)
    print('intents num :', intents_num)

    model = JointBertModel(slots_num, intents_num, bert_model_hub_path, sess,
                        num_bert_fine_tune_layers=10, is_bert=is_bert)


    print('train input shape :', train_input_ids.shape, train_input_ids[0:2])
    print('train_input_mask :', train_input_mask.shape, train_input_mask[0:2])
    print('train_segment_ids :', train_segment_ids.shape, train_segment_ids[0:2])
    print('train_tags :', train_tags.shape, train_tags[0:2])
    print('train_intents :', train_intents.shape, train_intents[0:2])


    print('training model ...')

    model.fit([train_input_ids, train_input_mask, train_segment_ids], [train_tags, train_intents],
    #          validation_data=([val_input_ids, val_input_mask, val_segment_ids, val_valid_positions], [val_tags, val_intents]),
            validation_data=None,
            epochs=epochs, batch_size=batch_size)

### saving
print('Saving ..')
if not os.path.exists(save_folder_path):
    os.makedirs(save_folder_path)
    print('Folder `%s` created' % save_folder_path)
model.save(save_folder_path)
with open(os.path.join(save_folder_path, 'tags_vectorizer.pkl'), 'wb') as handle:
    pickle.dump(tags_vectorizer, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open(os.path.join(save_folder_path, 'intents_label_encoder.pkl'), 'wb') as handle:
    pickle.dump(intents_label_encoder, handle, protocol=pickle.HIGHEST_PROTOCOL)


tf.compat.v1.reset_default_graph()