-
Notifications
You must be signed in to change notification settings - Fork 2
/
model_utils.py
160 lines (139 loc) · 5.84 KB
/
model_utils.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
import os
from text_token import _UNK, _PAD, _BOS, _EOS
import numpy as np
import torch
from torch import optim
def collate_fn_sg(batch):
'''
n_layers = len(batch[0][1])
is_inter = False
if is_inter:
encoder_input, decoder_labels, inter_labels = \
[], [[] for _ in range(n_layers)], [[] for _ in range(n_layers)]
else:
encoder_input, decoder_labels, refs, sf_data = [], [[] for _ in range(n_layers)], [], []
for data in batch:
encoder_input.append(data[0])
for idx in range(n_layers):
decoder_labels[idx].append(data[1][idx])
if is_inter:
inter_labels[idx].append(data[2][idx])
en_max_length = max([len(sent) for sent in encoder_input])
de_max_lengths = [
max([len(sent) for sent in labels]) for labels in decoder_labels]
de_lengths = [
sum(len(sent) for sent in labels) for labels in decoder_labels]
encoder_input = pad_sequences(encoder_input, en_max_length, 'pre')
for idx in range(n_layers):
decoder_labels[idx] = \
pad_sequences(decoder_labels[idx], de_max_lengths[idx], 'post')
for data in batch:
refs.append(pad_sequences(data[2], de_max_lengths[-1], 'post'))
sf_data.append(data[3])
if is_inter:
for idx in range(n_layers):
inter_labels[idx] = \
pad_sequences(inter_labels[idx], de_max_lengths[idx], 'post')
return encoder_input, decoder_labels, inter_labels, de_lengths
else:
# return encoder_input, decoder_labels, de_lengths
return encoder_input, decoder_labels, de_lengths, refs, sf_data
'''
encoder_input, decoder_label, refs, sf_data = [], [], [], []
for data in batch:
encoder_input.append(data[0])
decoder_label.append(data[1])
de_max_length = max([len(sent) for sent in decoder_label])
decoder_label = pad_sequences(decoder_label, de_max_length, 'post')
for data in batch:
refs.append(pad_sequences(data[2], de_max_length, 'post'))
sf_data.append(data[3])
return encoder_input, decoder_label, refs, sf_data
def collate_fn_sc(batch):
encoder_input, decoder_label, refs, sf_data = [], [], [], []
for data in batch:
encoder_input.append(data[1])
decoder_label.append(data[0])
en_max_length = max([len(sent) for sent in encoder_input])
encoder_input = pad_sequences(encoder_input, en_max_length, 'pre')
for data in batch:
refs.append(pad_sequences(data[2], en_max_length, 'pre'))
sf_data.append(data[3])
return encoder_input, decoder_label, refs, sf_data
def collate_fn_nl(batch):
encoder_input, decoder_label, refs, sf_data = [], [], [], []
for data in batch:
encoder_input.append([_BOS] + data[1][:-1])
decoder_label.append(data[1])
en_max_length = max([len(sent) for sent in encoder_input])
de_max_length = max([len(sent) for sent in decoder_label])
assert en_max_length == de_max_length
de_length = sum(len(sent) for sent in decoder_label)
encoder_input = pad_sequences(encoder_input, en_max_length, 'post')
decoder_label = pad_sequences(decoder_label, de_max_length, 'post')
for data in batch:
refs.append(pad_sequences(data[2], de_max_length, 'post'))
sf_data.append(data[3])
return encoder_input, decoder_label, refs, sf_data
# collated function for SF
def collate_fn_sf(batch):
'''
Warning: this function is not correctly implemented
'''
encoder_input, decoder_label, refs, sf_data = [], [], [], []
for data in batch:
encoder_input.append(data[0])
decoder_labels.append(data[0])
en_max_length = max([len(sent) for sent in encoder_input])
de_max_length = max([len(sent) for sent in decoder_label])
de_length = sum(len(sent) for sent in decoder_label)
encoder_input = pad_sequences(encoder_input, en_max_length, 'pre')
decoder_label = pad_sequences(decoder_label, de_max_length, 'post')
for data in batch:
refs.append(pad_sequences(data[2], de_max_length, 'post'))
sf_data.append(data[3])
return encoder_input, decoder_label, refs, sf_data
def pad_sequences(data, max_length, pad_type):
# print(data)
if _PAD != -1:
padded_data = np.full((len(data), max_length), _PAD)
else:
padded_data = np.full((len(data), max_length), _UNK)
if pad_type == "post":
for idx, d in enumerate(data):
padded_data[idx][:min(max_length, len(d))] = \
d[:min(max_length, len(d))]
elif pad_type == "pre":
for idx, d in enumerate(data):
padded_data[idx][max(0, max_length-len(d)):] = \
d[:min(max_length, len(d))]
return padded_data
# Model helper
def get_embeddings(vocab, embeddings_dir, embedding_dim):
embedding_file = os.path.join(
embeddings_dir, "glove.6B.{}d.txt".format(embedding_dim))
embeddings = torch.nn.Parameter(
torch.Tensor(torch.randn(len(vocab), embedding_dim)))
with open(embedding_file, 'r') as file:
for line in file:
data = line.strip().split(' ')
word, emb = \
data[0], torch.Tensor(np.array(list(map(float, data[1:]))))
if word not in vocab:
continue
embeddings.data[vocab[word]] = torch.Tensor(emb)
return embeddings
def get_device(device=None):
if device is not None:
return torch.device(device)
return torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
def build_optimizer(optimizer, parameters, learning_rate):
if optimizer == "Adam":
return optim.Adam(
parameters, lr=learning_rate)
elif optimizer == "RMSprop":
return optim.RMSprop(
parameters, lr=learning_rate)
elif optimizer == "SGD":
return optim.SGD(
parameters, lr=learning_rate)