-
Notifications
You must be signed in to change notification settings - Fork 4
/
training_split.py
200 lines (149 loc) · 5.87 KB
/
training_split.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 __future__ import print_function
from __future__ import absolute_import
import os
import sys
import logging
import numpy as np
from collections import defaultdict
from preprocess import serialize, error_dict
from sklearn.cross_validation import train_test_split
from keras.layers.embeddings import Embedding
from keras.models import Sequential
from seq2seq.models import SimpleSeq2seq
from keras.layers.core import Dropout
# from keras.layers import Input
# from keras.layers.recurrent import LSTM
# from keras.layers.core import Dense, Dropout
# from keras.layers.wrappers import TimeDistributed
# from keras.models import Model
from sklearn.metrics import precision_score, recall_score, accuracy_score, f1_score
bucket_size = 10
random_state = 0
embedding_dim = 200
batch_size = 20
nb_epoch = 3
hidden_dim = 100
def encode(ret_text, ret_label):
vocab = defaultdict(int)
for i in range(len(ret_text)):
for j in range(len(ret_text[i])):
vocab[ret_text[i][j]] = vocab[ret_text[i][j]] + 1
# for debug
# ret_text = ret_text[:200]
# ret_label = ret_label[:200]
# train, test split
train_text, test_text, train_label, test_label = train_test_split(ret_text, ret_label,
random_state=random_state, test_size=0.1)
# print(len(train_text), len(test_text))
logging.info('training samples: %d' % len(train_text))
logging.info('testing samples: %d' % len(test_text))
word_idx_map = dict()
i = 1
for word in vocab.keys():
word_idx_map[word] = i
i = i + 1
X_train, y_train, X_test, y_test = [], [], [], []
for i in range(len(train_text)):
cur_line = []
cur_label = []
move_point = 1
for j in range(len(train_text[i])):
cur_line.append(word_idx_map[train_text[i][j]])
cur_label.append(train_label[i][j])
move_point = move_point + 1
if move_point > bucket_size:
X_train.append(cur_line)
y_train.append(cur_label)
cur_line = []
cur_label = []
move_point = 1
if len(cur_line) > 0:
for k in range(len(cur_line), 10):
cur_line.append(0)
cur_label.append(0)
X_train.append(cur_line)
y_train.append(cur_label)
# y_train = np.array(y_train)
new_y_train = np.zeros((len(y_train), bucket_size, len(error_dict.keys())+1))
for i in range(len(y_train)):
for j in range(len(y_train[i])):
new_y_train[i, j, y_train[i][j]] = 1
# print(new_y_train[i][j])
X_train = np.array(X_train)
for i in range(len(test_text)):
cur_line = []
cur_label = []
move_point = 1
for j in range(len(test_text[i])):
cur_line.append(word_idx_map[test_text[i][j]])
cur_label.append(test_label[i][j])
move_point = move_point + 1
if move_point > bucket_size:
X_test.append(cur_line)
y_test.append(cur_label)
cur_line = []
cur_label = []
move_point = 1
if len(cur_line) > 0:
for k in range(len(cur_line), 10):
cur_line.append(0)
cur_label.append(0)
X_test.append(cur_line)
y_test.append(cur_label)
X_test = np.array(X_test)
y_test = np.array(y_test)
return X_train, new_y_train, X_test, y_test, vocab
def decode(X_test, y_test, predict_array):
y_pred = []
y_true = []
for i in range(len(X_test)):
for j in range(len(X_test[i])):
if X_test[i][j] != 0:
y_true.append(y_test[i][j])
print(predict_array[i][j], np.argmax(predict_array[i][j]))
cur_pred = np.argmax(predict_array[i][j])
y_pred.append(cur_pred)
print(y_true, y_pred)
# print(len(y_true), len(y_pred))
precision = precision_score(y_true, y_pred, average='micro')
recall = recall_score(y_true, y_pred, average='micro')
accuracy = accuracy_score(y_true, y_pred)
print(precision, recall, accuracy)
if __name__ == '__main__':
program = os.path.basename(sys.argv[0])
logger = logging.getLogger(program)
logging.basicConfig(format='%(asctime)s: %(levelname)s: %(message)s')
logging.root.setLevel(level=logging.INFO)
logger.info(r"running %s" % ''.join(sys.argv))
input_file = os.path.join('data', 'CGED16_HSK_Train_All.txt')
ret_id, ret_text, ret_label = serialize(input_file)
X_train, y_train, X_test, y_test, vocab = encode(ret_text, ret_label)
feature_length = len(vocab.keys()) + 1
model = Sequential()
# model.add(Input(shape=(bucket_size, ), dtype='int32'))
model.add(Embedding(feature_length, embedding_dim, input_length=bucket_size))
model.add(Dropout(0.25))
seq2seq = SimpleSeq2seq(
input_dim=embedding_dim,
input_length=bucket_size,
hidden_dim=100,
output_dim=len(error_dict.keys()) + 1,
output_length=bucket_size,
depth=3
)
model.add(seq2seq)
# sequence = Input(shape=(bucket_size, ), dtype='int32')
# embedded = Embedding(input_dim=feature_length, output_dim=embedding_dim, input_length=bucket_size) (sequence)
# embedded = Dropout(0.25) (embedded)
# # encoder
# encoder = LSTM(hidden_dim, return_sequences=True) (embedded)
# encoder = Dropout(0.25) (encoder)
# decoder = LSTM(hidden_dim, return_sequences=True) (encoder)
# decoder = Dropout(0.25) (decoder)
# output = TimeDistributed(Dense(1)) (decoder)
# model = Model(input=sequence, output=output)
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')
model.fit(X_train, y_train, batch_size=batch_size, nb_epoch=nb_epoch)
predict_array = model.predict(X_test, batch_size=batch_size)
print(predict_array.shape)
decode(X_test, y_test, predict_array)