-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
164 lines (116 loc) · 6.92 KB
/
main.py
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
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
import numpy as np
import gensim
import pickle
from model import NERModel
from train import train
import utils.evaluate as evaluate
import utils.prepare_data as prepare_data
from utils import radam
from config.params import *
if __name__ == '__main__':
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
torch.manual_seed(0)
print(device)
whole_data_path = 'data/combined/whole.txt'
train_data_path = 'data/combined/train.txt'
dev_data_path = 'data/combined/dev.txt'
test_data_path = 'data/combined/digitoday.2015.test.txt'
wiki_data_path = 'data/combined/wikipedia.test.txt'
whole_data_morph_path = 'utils/subword_segmentation/output/segmented/whole_vocab_segmented.txt'
whole_data = prepare_data.load_data(whole_data_path)
train_data = prepare_data.load_data(train_data_path)
dev_data = prepare_data.load_data(dev_data_path)
test_data = prepare_data.load_data(test_data_path)
wiki_data = prepare_data.load_data(wiki_data_path)
whole_data_morphs = prepare_data.load_data_morphs(whole_data_morph_path)
# convert to lower case
whole_data = prepare_data.to_lower(whole_data)
train_data = prepare_data.to_lower(train_data)
dev_data = prepare_data.to_lower(dev_data)
test_data = prepare_data.to_lower(test_data)
wiki_data = prepare_data.to_lower(wiki_data)
# remove punctuation
whole_data = prepare_data.remove_punct(whole_data)
train_data = prepare_data.remove_punct(train_data)
dev_data = prepare_data.remove_punct(dev_data)
test_data = prepare_data.remove_punct(test_data)
wiki_data = prepare_data.remove_punct(wiki_data)
# add <start> and <end> token to each sentence
whole_data = prepare_data.add_start_end_sentence_tokens(whole_data)
train_data = prepare_data.add_start_end_sentence_tokens(train_data)
dev_data = prepare_data.add_start_end_sentence_tokens(dev_data)
test_data = prepare_data.add_start_end_sentence_tokens(test_data)
wiki_data = prepare_data.add_start_end_sentence_tokens(wiki_data)
print('Loading embeddings...')
# embeddings = gensim.models.KeyedVectors.load_word2vec_format('data/embeddings/fin-word2vec.bin', binary=True, limit=500000)
embeddings = gensim.models.fasttext.load_facebook_vectors('data/embeddings/cc.fi.300.bin')
print('Finished loading embeddings')
word2idx, idx2word, tag2idx, idx2tag, char2idx, idx2char = prepare_data.encode_data(whole_data)
morph2idx, idx2morph = prepare_data.encode_data_morphs(whole_data_morphs)
word2morph = prepare_data.word_to_morph(whole_data_morphs)
with open('weights/char_dict_lower.pkl', 'wb') as f:
pickle.dump(char2idx, f, pickle.HIGHEST_PROTOCOL)
with open('weights/morph_dict_lower.pkl', 'wb') as f:
pickle.dump(morph2idx, f, pickle.HIGHEST_PROTOCOL)
indexed_data_train = prepare_data.data_to_idx(train_data, word2idx, embeddings)
indexed_tag_train = prepare_data.tag_to_idx(train_data, tag2idx)
indexed_char_train = prepare_data.char_to_idx(train_data, char2idx)
indexed_morph_train = prepare_data.morph_to_idx(train_data, morph2idx, word2morph)
data_train = prepare_data.combine_data(indexed_data_train, indexed_tag_train, indexed_char_train, indexed_morph_train, MAX_SEQ_LENGTH)
indexed_data_dev = prepare_data.data_to_idx(dev_data, word2idx, embeddings)
indexed_tag_dev = prepare_data.tag_to_idx(dev_data, tag2idx)
indexed_char_dev = prepare_data.char_to_idx(dev_data, char2idx)
indexed_morph_dev = prepare_data.morph_to_idx(dev_data, morph2idx, word2morph)
data_dev = prepare_data.combine_data(indexed_data_dev, indexed_tag_dev, indexed_char_dev, indexed_morph_dev, MAX_SEQ_LENGTH)
indexed_data_test = prepare_data.data_to_idx(test_data, word2idx, embeddings)
indexed_tag_test = prepare_data.tag_to_idx(test_data, tag2idx)
indexed_char_test = prepare_data.char_to_idx(test_data, char2idx)
indexed_morph_test = prepare_data.morph_to_idx(test_data, morph2idx, word2morph)
data_test = prepare_data.combine_data(indexed_data_test, indexed_tag_test, indexed_char_test, indexed_morph_test, MAX_SEQ_LENGTH)
indexed_data_wiki = prepare_data.data_to_idx(wiki_data, word2idx, embeddings)
indexed_tag_wiki = prepare_data.tag_to_idx(wiki_data, tag2idx)
indexed_char_wiki = prepare_data.char_to_idx(wiki_data, char2idx)
indexed_morph_wiki = prepare_data.morph_to_idx(wiki_data, morph2idx, word2morph)
data_wiki = prepare_data.combine_data(indexed_data_wiki, indexed_tag_wiki, indexed_char_wiki, indexed_morph_wiki, MAX_SEQ_LENGTH)
data_train = prepare_data.remove_extra(data_train, batch_size)
data_dev = prepare_data.remove_extra(data_dev, batch_size)
pairs_batch_train = DataLoader(dataset=data_train,
batch_size=batch_size,
shuffle=True,
collate_fn=prepare_data.collate,
pin_memory=True)
pairs_batch_dev = DataLoader(dataset=data_dev,
batch_size=batch_size,
shuffle=True,
collate_fn=prepare_data.collate,
pin_memory=True)
# initialize the model
model = NERModel(word_embedding_dim, char_embedding_dim, morph_embedding_dim, word_hidden_size, char_hidden_size, morph_hidden_size,
len(char2idx), len(morph2idx), len(tag2idx)+1, word_num_layers, char_num_layers, morph_num_layers, dropout_prob).to(device)
model.train()
criterion = nn.NLLLoss()
optimizer = radam.RAdam(model.parameters(), lr=learning_rate)
print(model)
total_trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
print('The number of trainable parameters is: %d' % (total_trainable_params))
# train the model
if skip_training == False:
train(model, word_num_layers, char_num_layers, morph_num_layers, num_epochs, pairs_batch_train, pairs_batch_dev, word_hidden_size,
char_hidden_size, morph_hidden_size, batch_size, criterion, optimizer, patience, device)
model.load_state_dict(torch.load('weights/model_lower.pt'))
else:
model.load_state_dict(torch.load('weights/model_lower.pt'))
model.eval()
batch_size = 1
print('\nTEST DATA \n')
all_predicted, all_true = evaluate.get_predictions(data_test, model, word_num_layers, char_num_layers, morph_num_layers, word_hidden_size,
char_hidden_size, morph_hidden_size, batch_size, device)
evaluate.print_scores(all_predicted, all_true, tag2idx)
print('\nWIKI DATA \n')
all_predicted, all_true = evaluate.get_predictions(data_wiki, model, word_num_layers, char_num_layers, morph_num_layers, word_hidden_size,
char_hidden_size, morph_hidden_size, batch_size, device)
evaluate.print_scores(all_predicted, all_true, tag2idx)