-
Notifications
You must be signed in to change notification settings - Fork 0
/
util.py
200 lines (147 loc) · 6.45 KB
/
util.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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
from collections import defaultdict
from IPython.display import clear_output
import matplotlib.pyplot as plt
import numpy as np
import torch
import torch.nn.functional as F
from torch.utils.data import Dataset
from torch.autograd import Variable
from torchtext.data import get_tokenizer
tokenizer = get_tokenizer("basic_english")
class CorpusDictionary:
def __init__(self, texts, unknown_symbol="UNK", pad_symbol="PAD"):
tokenized_texts = list(map(tokenizer, texts))
self.dictionary = self._create_dictionary(tokenized_texts)
self.n_texts = len(texts)
self.unknown_symbol = unknown_symbol
self.pad_symbol = pad_symbol
self.word_to_idx, self.idx_to_word = self._create_indexers(unknown_symbol, pad_symbol)
def _create_dictionary(self, tokenized_texts):
dictionary = defaultdict(int)
for tokens in tokenized_texts:
tokens = set(tokens)
for token in tokens:
dictionary[token] += 1
return dictionary
def _create_indexers(self, unknown_symbol, pad_symbol):
word_to_idx = {word: idx for word, idx in zip(self.dictionary.keys(), range(2, len(self.dictionary) + 2))}
word_to_idx[pad_symbol] = 0
word_to_idx[unknown_symbol] = 1
idx_to_word = {idx: word for word, idx in word_to_idx.items()}
return word_to_idx, idx_to_word
def get_frequencies(self):
return sorted(
[(word, frequency / self.n_texts) for word, frequency in self.dictionary.items()],
key=lambda x: x[1]
)
def truncate_dictionary(self, min_frequency=0.0, max_frequency=1.0):
self.dictionary = {
word: frequency
for word, frequency in self.dictionary.items()
if frequency / self.n_texts >= min_frequency
and frequency / self.n_texts <= max_frequency
}
self.word_to_idx, self.idx_to_word = self._create_indexers(self.unknown_symbol, self.pad_symbol)
def transform(self, texts):
tokenized_texts = list(map(tokenizer, texts))
return [
[
self.word_to_idx.get(
token,
self.word_to_idx[self.unknown_symbol]
) for token in tokens
] for tokens in tokenized_texts
]
class PaddedTextVectorDataset(Dataset):
def __init__(self, texts, target, corpus_dict=None, emb=None, max_vector_len=50):
self.max_vector_len = max_vector_len
vectors = self._get_vectors(texts, corpus_dict, emb)
self.lengths = list(map(
lambda x: len(x) if len(x) <= max_vector_len else max_vector_len,
vectors
))
self.vectors = list(map(self.pad_data, vectors))
self.target = list(target)
assert len(texts) == len(target), "Texts len != target len: {} != {}".format(len(texts), len(target))
def _get_vectors(self, texts, corpus_dict, emb):
if corpus_dict is not None:
assert emb is None, "Can't provide both corpus_dict and pretrained embeddings"
return corpus_dict.transform(texts)
if emb is not None:
assert corpus_dict is None, "Can't provide both corpus_dict and pretrained embeddings"
tokenized_texts = list(map(tokenizer, texts))
vectors = [[emb.stoi[token] for token in tokens if token in emb.stoi] for tokens in tokenized_texts]
return [vector if len(vector) > 0 else [0] for vector in vectors]
raise ValueError("Should provide one of: corpus_dict, pretrained embeddings")
def __len__(self):
return len(self.target)
def __getitem__(self, idx):
X = np.asarray(self.vectors[idx])
vector_len = np.asarray(self.lengths[idx])
y = np.asarray(self.target[idx]).astype(float)
return X, y, vector_len
def pad_data(self, s):
padded = np.zeros((self.max_vector_len,), dtype=np.int64)
if len(s) > self.max_vector_len:
padded[:] = s[:self.max_vector_len]
else:
padded[:len(s)] = s
return padded
def sort_batch(X, y, lengths):
lengths, indx = lengths.sort(dim=0, descending=True)
X = X[indx]
y = y[indx]
return X.transpose(0, 1), y, lengths
def fit(model, train_dl, test_dl, loss_fn, opt, epochs=3):
train_losses = []
test_losses = []
for epoch in range(epochs):
model.train()
total_loss_train = 0
total_loss_test = 0
for X, y, lengths in iter(train_dl):
X, y, lengths = sort_batch(X, y, lengths)
X = Variable(X)
y = Variable(y)
lengths = lengths.numpy()
opt.zero_grad()
pred = model(X, lengths)
loss = loss_fn(pred, y)
loss.backward()
opt.step()
total_loss_train += loss.item()
train_loss = total_loss_train / len(train_dl)
train_losses.append(train_loss)
if epoch % 5 == 0:
model.eval()
for X, y, lengths in test_dl:
X, y, lengths = sort_batch(X, y, lengths)
X = Variable(X)
y = Variable(y)
pred = model(X, lengths.numpy())
loss = loss_fn(pred, y)
total_loss_test += loss.item()
test_loss = total_loss_test / len(test_dl)
test_losses.append(test_loss)
clear_output(wait=True)
print("Train loss:\t{:.4f}\nVal loss:\t{:.4f}".format(train_loss, test_losses[-1]))
plt.figure(figsize=(8, 4))
plt.plot(range(epoch + 1), train_losses, label='train', marker='o')
plt.plot(range(0, epoch + 1, 5), test_losses, label='test', marker='o')
plt.legend()
plt.title("Training loss")
plt.show()
def predict(model, dl):
y_true_val, y_pred_val, y_pred_proba = [], [], []
model.eval()
for X, y, lengths in dl:
X, y, lengths = sort_batch(X, y, lengths)
X = Variable(X)
y = Variable(y)
pred = model(X, lengths.numpy())
prob = F.softmax(pred, dim=1)
pred_idx = torch.max(pred, 1)[1]
y_true_val += list(y.cpu().data.numpy())
y_pred_val += list(pred_idx.cpu().data.numpy())
y_pred_proba += list(prob.cpu().data.numpy())
return np.asarray(y_true_val), np.asarray(y_pred_val), np.asarray(y_pred_proba)