-
Notifications
You must be signed in to change notification settings - Fork 37
/
Copy pathquora_dssm.py
212 lines (160 loc) · 7.61 KB
/
quora_dssm.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
201
202
203
204
205
206
207
208
209
210
211
212
#coding=utf-8
import sys
import numpy as np
import tensorflow as tf
from sklearn.datasets import load_svmlight_file
from sklearn.model_selection import KFold
from sklearn.utils import shuffle
import config
from DSSM.dssm import DSSM
from helper import tools
def get_words(sentences):
words = []
for one_sen in sentences:
one_sen = one_sen.strip().split()
one_sen = map(lambda x: x.strip(), one_sen)
words += one_sen
return words
def quora_dssm(train_input_file, test_input_file):
'''
测试函数
:return:
'''
seed = 2222
'''
train_ori = pd.read_csv(train_input_file)
test = pd.read_csv(test_input_file)#, nrows=1001)
test['is_duplicate'] = 0
# train_ori = train_ori[:1000]
# test = test_ori[:]
print train_ori['is_duplicate'].value_counts()
q = ['question1', 'question2']
words = []
for one_q in q:
train_ori[one_q] = train_ori[one_q].astype(str)
test[one_q] = test[one_q].astype(str)
'''
#wordhash = WordHash(words, load_from_file=True, \
# dump_to_file=True, file='result/n_gram_term_index_mapping.pkl')
#print 'Load n_gram_term_index_mapping.pkl done'
#sys.stdout.flush()
#train_ori_q1 = wordhash.get_n_gram_count(train_ori[q[0]].values, is_dump=True, dump_file='result/train_q1_ngram_counting_matrix.pkl')
#train_ori_q2 = wordhash.get_n_gram_count(train_ori[q[1]].values, is_dump=True, dump_file='result/train_q2_ngram_counting_matrix.pkl')
'''
with open('result/train_q1_ngram_counting_matrix.pkl', 'rb') as fr:
train_ori_q1 = pickle.load(fr)
with open('result/train_q2_ngram_counting_matrix.pkl', 'rb') as fr:
train_ori_q2 = pickle.load(fr)
print 'Get train origin sparse matrix done'
sys.stdout.flush()
'''
'''
y = train_ori['is_duplicate'].values[:]
y_t = test['is_duplicate'].values[:]
del train_ori
test_q1 = pd.read_pickle('result/test_q1_ngram_counting_matrix.pkl')
test_q2 = pd.read_pickle('result/test_q2_ngram_counting_matrix.pkl')
#test_q1 = wordhash.get_n_gram_count(test[q[0]].values, is_dump=True, dump_file='result/test_q1_ngram_counting_matrix.pkl')
#test_q2 = wordhash.get_n_gram_count(test[q[1]].values, is_dump=True, dump_file='result/test_q2_ngram_counting_matrix.pkl')
del test
print 'Get test origin sparse matrix done'
sys.stdout.flush()
'''
'''
X = sps.hstack(
[train_ori_q1, train_ori_q2]
).tocsr()
X_t = sps.hstack(
[test_q1, test_q2]
).tocsr()
'''
'''
Get origin train and test svm format file.
'''
#dump_svmlight_file(X, y, 'result/train_ori_n_gram_counting_sparse_matrix.svm')
#dump_svmlight_file(X_t, y_t, 'result/test_n_gram_counting_sparse_matrix.svm')
#X, y, X_test, y_test = load_svmlight_files(['result/train_ori_n_gram_counting_sparse_matrix.svm', 'result/test_n_gram_counting_sparse_matrix.svm']) # 注意load_svmlight_file的shape是根据推断来的，可能导致不一致
#X = normalize(X, axis=0)
#X_test = normalize(X_test, axis=0)
#dump_svmlight_file(X, y, 'result/train_ori_n_gram_counting_sparse_matrix.norm.svm')
#dump_svmlight_file(X_test, y_test, 'result/test_n_gram_counting_sparse_matrix.norm.svm')
X, y = load_svmlight_file('result/train_ori_n_gram_counting_sparse_matrix.norm.mini.svm', n_features=111166)
print X.shape
used_as_train = X.shape[0]/10
X = X[:used_as_train]
y = y[:used_as_train]
print y[:10]
#dump_svmlight_file(X, y, 'result/train_ori_n_gram_counting_sparse_matrix.norm.mini.svm')
#print 'X_train shape: ', X.shape, ' X_test shape: ', X_test.shape
print 'Load done'
sys.stdout.flush()
skf = KFold(n_splits=5, shuffle=True, random_state=seed).split(X)
for ind_tr, ind_te in skf:
X_train = X[ind_tr]
y_train = y[ind_tr]
X_valid = X[ind_te]
y_valid = y[ind_te]
break
X_train, y_train = tools.oversample(X_train.tocsr(), y_train, p=0.165)
X_valid, y_valid = tools.oversample(X_valid.tocsr(), y_valid, p=0.165)
X_train, y_train = shuffle(X_train, y_train, random_state=seed)
#dump_svmlight_file(X_train, y_train, 'result/oversample_train_n_gram_counting_sparse_matrix.svm')
#dump_svmlight_file(X_valid, y_valid, 'result/oversample_valid_n_gram_counting_sparse_matrix.svm')
#print 'Dump to svm format done.'
'''
for _ in q:
train_ori[_] = train_ori[_].astype(str)
test[_] = test_ori[_].astype(str)
words += get_words(train_ori[_].values)
words += get_words(test[_].values)
print 'Sum words: ', len(words), ' sum diff words: ', len(set(words))
wordhash = WordHash(words, load_from_file=True, load_file='n_gram_term_index_mapping.pkl', \
dump_to_file=True, dump_file='n_gram_term_index_mapping.pkl')
split_point = int(0.7 * len(train_ori))
train = train_ori[:split_point]
valid = train_ori[split_point:]
train_q1 = wordhash.get_n_gram_count(train[q[0]].values, is_dump=True, dump_file='result/train_q1_ngram_counting_matrix.pkl')
train_q2 = wordhash.get_n_gram_count(train[q[1]].values, is_dump=True, dump_file='result/train_q2_ngram_counting_matrix.pkl')
train_label = train['is_duplicate'].values
valid_q1 = wordhash.get_n_gram_count(valid[q[0]].values, is_dump=True, dump_file='result/valid_q1_ngram_counting_matrix.pkl')
valid_q2 = wordhash.get_n_gram_count(valid[q[1]].values, is_dump=True, dump_file='result/valid_q2_ngram_counting_matrix.pkl')
valid_label = valid['is_duplicate'].values
test_q1 = wordhash.get_n_gram_count(test[q[0]].values, is_dump=True, dump_file='result/test_q1_ngram_counting_matrix.pkl')
test_q2 = wordhash.get_n_gram_count(test[q[1]].values, is_dump=True, dump_file='result/test_q2_ngram_counting_matrix.pkl')
test_label = test['is_duplicate'].values
'''
print 'train shape: ', X_train.shape, 'valid shape: ', X_valid.shape
#print 'test shape: ', X_test.shape
n_gram_size = X_train.shape[1]
#n_gram_size = X_test.shape[1]
sys.stdout.flush()
with tf.Graph().as_default():
tf.set_random_seed(1)
model = DSSM(hash_tokens_nums=n_gram_size/2, dnn_layer_nums=2, dnn_hidden_node_nums=288, feature_nums=64, batch_size=X_train.shape[0], neg_nums=0, learning_rate=0.001, max_epochs=400, loss_kind='log_loss', w_init=1,save_model_path='result/save-model', mlp_hidden_node_nums=16, mlp_layer_nums=100,input_is_sparse=True)
sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
np.random.seed(1)
# query = np.random.rand(500, 30000)
# doc = np.random.rand(500, 30000)
# label = np.array([1, 0, 0, 0, 0] * 100)
# model.set_positive_weights([1]*500)
#print query
#print doc
#print label
X_train_q1 = X_train[:, :n_gram_size/2]
X_train_q2 = X_train[:, n_gram_size/2:]
X_valid_q1 = X_valid[:, :n_gram_size/2]
X_valid_q2 = X_valid[:, n_gram_size/2:]
#X_test_q1 = X_test[:, :n_gram_size/2]
#X_test_q2 = X_test[:, n_gram_size/2:]
losses = model.fit(sess, X_train_q1, X_train_q2, y_train, X_valid_q1, X_valid_q2, y_valid, load_model=False)
'''
print 'Start to test. '
test['is_duplicate'] = model.predict(sess, X_test_q1, X_test_q2, y_test, is_sparse=True)
test[['test_id', 'is_duplicate']].to_csv('result/out.csv', index=False)
'''
if __name__ == '__main__':
train_file = config.train_file
test_file = config.test_file
quora_dssm(train_file, test_file)