forked from sunilitggu/DDI-extraction-through-LSTM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain_train_val.py
370 lines (284 loc) · 10.6 KB
/
main_train_val.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
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
from sklearn.metrics import precision_score
from sklearn.metrics import recall_score
from sklearn.metrics import f1_score
from sklearn.metrics import confusion_matrix
import numpy as np
import sklearn as sk
import random
import csv
import re
import collections
import pickle
import sys
sys.path.append("source")
from utils import *
from joint_ablstm import *
#from rnn_train import *
#from cnn_train import *
#from att_rnn import *
embSize = 100
d1_emb_size=10
d2_emb_size=10
type_emb_size=10
numfilter = 200
#out_file = sys.argv[1]
#sent_out = sys.argv[2]
out_file = 'results/multi_results.txt'
sent_out = 'results/multi_sents_'
num_epochs = 18
#N = 4
check_point = [4,7,10,13,17]
batch_size=200
reg_para = 0.001
drop_out = 1.0
ftrain = "dataset/ddi/neg_filtered/train_data95.txt"
fval = "dataset/ddi/neg_filtered/train_data05.txt"
ftest = "dataset/ddi/neg_filtered/test_data.txt"
#wefile = "/home/sunil/embeddings/cbow_300d_gvkcorpus.txt"
wefile = "/home/sunil/embeddings/glove_100d_w9_pubmed.txt"
Tr_sent_contents, Tr_entity1_list, Tr_entity2_list, Tr_sent_lables = dataRead(ftrain)
Tr_word_list, Tr_d1_list, Tr_d2_list, Tr_type_list = makeFeatures(Tr_sent_contents, Tr_entity1_list, Tr_entity2_list)
V_sent_contents, V_entity1_list, V_entity2_list, V_sent_lables = dataRead(fval)
V_word_list, V_d1_list, V_d2_list, V_type_list = makeFeatures(V_sent_contents, V_entity1_list, V_entity2_list)
Te_sent_contents, Te_entity1_list, Te_entity2_list, Te_sent_lables = dataRead(ftest)
Te_word_list, Te_d1_list, Te_d2_list, Te_type_list = makeFeatures(Te_sent_contents, Te_entity1_list, Te_entity2_list)
print "train_size", len(Tr_word_list)
print "val_size", len(V_word_list)
print "test_size", len(Te_word_list)
train_sent_lengths, val_sent_lengths, test_sent_lengths = findSentLengths([Tr_word_list, V_word_list, Te_word_list])
sentMax = max(train_sent_lengths + val_sent_lengths + test_sent_lengths)
print "max sent length", sentMax
train_sent_lengths = np.array(train_sent_lengths, dtype='int32')
val_sent_lengths = np.array(train_sent_lengths, dtype='int32')
test_sent_lengths = np.array(test_sent_lengths, dtype='int32')
label_dict = {'false':0, 'advise': 1, 'mechanism': 2, 'effect': 3, 'int': 4}
#label_dict = {'false':0, 'true':1}
word_dict = makeWordList([Tr_word_list, V_word_list, Te_word_list])
d1_dict = makeDistanceList([Tr_d1_list, V_d1_list, Te_d1_list])
d2_dict = makeDistanceList([Tr_d2_list, V_d2_list, Te_d2_list])
type_dict = makeDistanceList([Tr_type_list, V_type_list, Te_type_list])
print "word dictonary length", len(word_dict)
# Word Embedding
wv = readWordEmb(word_dict, wefile, embSize)
# Mapping Train
W_train = mapWordToId(Tr_word_list, word_dict)
d1_train = mapWordToId(Tr_d1_list, d1_dict)
d2_train = mapWordToId(Tr_d2_list, d2_dict)
T_train = mapWordToId(Tr_type_list,type_dict)
Y_t = mapLabelToId(Tr_sent_lables, label_dict)
Y_train = np.zeros((len(Y_t), len(label_dict)))
for i in range(len(Y_t)):
Y_train[i][Y_t[i]] = 1.0
#Mapping Validation
W_val = mapWordToId(V_word_list, word_dict)
d1_val = mapWordToId(V_d1_list, d1_dict)
d2_val = mapWordToId(V_d2_list, d2_dict)
T_val = mapWordToId(V_type_list,type_dict)
Y_t = mapLabelToId(V_sent_lables, label_dict)
Y_val = np.zeros((len(Y_t), len(label_dict)))
for i in range(len(Y_t)):
Y_val[i][Y_t[i]] = 1.0
# Mapping Test
W_test = mapWordToId(Te_word_list, word_dict)
d1_test = mapWordToId(Te_d1_list, d1_dict)
d2_test = mapWordToId(Te_d2_list, d2_dict)
T_test = mapWordToId(Te_type_list, type_dict)
Y_t = mapLabelToId(Te_sent_lables, label_dict)
Y_test = np.zeros((len(Y_t), len(label_dict)))
for i in range(len(Y_t)):
Y_test[i][Y_t[i]] = 1.0
#padding
W_train, d1_train, d2_train, T_train, W_val, d1_val, d2_val, T_val, W_test, d1_test, d2_test, T_test = paddData([W_train, d1_train, d2_train, T_train, W_val, d1_val, d2_val, T_val, W_test, d1_test, d2_test, T_test], sentMax)
print "train", len(W_train)
print "test", len(W_test)
with open('train_test_rnn_data.pickle', 'wb') as handle:
pickle.dump(W_train, handle)
pickle.dump(d1_train, handle)
pickle.dump(d2_train, handle)
pickle.dump(T_train, handle)
pickle.dump(Y_train, handle)
pickle.dump(train_sent_lengths, handle)
pickle.dump(W_val, handle)
pickle.dump(d1_val, handle)
pickle.dump(d2_val, handle)
pickle.dump(T_val, handle)
pickle.dump(Y_val, handle)
pickle.dump(val_sent_lengths, handle)
pickle.dump(W_test, handle)
pickle.dump(d1_test, handle)
pickle.dump(d2_test, handle)
pickle.dump(T_test, handle)
pickle.dump(Y_test, handle)
pickle.dump(test_sent_lengths, handle)
pickle.dump(wv, handle)
pickle.dump(word_dict, handle)
pickle.dump(d1_dict, handle)
pickle.dump(d2_dict, handle)
pickle.dump(type_dict, handle)
pickle.dump(label_dict, handle)
pickle.dump(sentMax, handle)
"""
with open('train_test_rnn_data.pickle', 'rb') as handle:
W_train = pickle.load(handle)
d1_train= pickle.load(handle)
d2_train= pickle.load(handle)
T_train = pickle.load(handle)
Y_train = pickle.load(handle)
train_sent_lengths = pickle.load(handle)
W_val = pickle.load(handle)
d1_val = pickle.load(handle)
d2_val = pickle.load(handle)
T_val = pickle.load(handle)
Y_val = pickle.load(handle)
val_sent_lengths = pickle.load(handle)
W_test = pickle.load(handle)
d1_test = pickle.load(handle)
d2_test = pickle.load(handle)
T_test = pickle.load(handle)
Y_test = pickle.load(handle)
test_sent_lengths = pickle.load(handle)
wv = pickle.load(handle)
word_dict= pickle.load(handle)
d1_dict = pickle.load(handle)
d2_dict = pickle.load(handle)
type_dict = pickle.load(handle)
label_dict = pickle.load(handle)
sentMax = pickle.load(handle)
"""
#vocabulary size
word_dict_size = len(word_dict)
d1_dict_size = len(d1_dict)
d2_dict_size = len(d2_dict)
type_dict_size = len(type_dict)
label_dict_size = len(label_dict)
rev_word_dict = makeWordListReverst(word_dict)
rev_label_dict = {0:'false', 1:'advise', 2:'mechanism', 3:'effect', 4:'int'}
fp = open(out_file, 'a+') # keep precision recall
fsent = open(sent_out, 'w') # keep sentence and its results
print 'drop_out, reg_rate', drop_out, reg_para
rnn = RNN_Relation(label_dict_size, # output layer size
word_dict_size, # word embedding size
d1_dict_size, # position embedding size
d2_dict_size, # position embedding size
type_dict_size, # type emb. size
sentMax, # length of sentence
wv, # word embedding
d1_emb_size=d1_emb_size, # emb. length
d2_emb_size=d2_emb_size,
type_emb_size=type_emb_size,
num_filters=numfilter, # number of hidden nodes in RNN
w_emb_size=embSize, # dim. word emb
l2_reg_lambda=reg_para # l2 reg
)
train_len = len(W_train)
loss_list = []
test_res = []
val_res = []
fscore_val = []
fscore_test = []
def test_step(W, sent_lengths, d1, d2, T, Y):
n = len(W)
# print 'n',n
ra = n/batch_size
samples = []
for i in range(ra):
samples.append(range(batch_size*i, batch_size*(i+1)))
samples.append(range(batch_size*(i+1), n))
acc = []
pred = []
for i in samples:
p,a = rnn.test_step(W[i], sent_lengths[i], d1[i], d2[i], T[i], Y[i])
# acc.extend(a)
pred.extend(p)
# print 'pred', len(pred)
return pred, acc
num_batches_per_epoch = int(train_len/batch_size) + 1
iii = 0 #Check point number
for epoch in range(num_epochs):
shuffle_indices = np.random.permutation(np.arange(train_len))
W_tr = W_train[shuffle_indices]
d1_tr = d1_train[shuffle_indices]
d2_tr = d2_train[shuffle_indices]
T_tr = T_train[shuffle_indices]
Y_tr = Y_train[shuffle_indices]
S_tr = train_sent_lengths[shuffle_indices]
loss_epoch = 0.0
for batch_num in range(num_batches_per_epoch):
start_index = batch_num*batch_size
end_index = min((batch_num + 1) * batch_size, train_len)
loss = rnn.train_step(W_tr[start_index:end_index], S_tr[start_index:end_index], d1_tr[start_index:end_index],
d2_tr[start_index:end_index], T_tr[start_index:end_index], Y_tr[start_index:end_index], drop_out)
loss_epoch += loss
print loss_epoch
loss_list.append(round(loss_epoch, 5) )
# if (epoch%N) == 0:
if epoch in check_point:
iii += 1
saver = tf.train.Saver()
path = saver.save(rnn.sess, 'saved_models/model_'+str(iii)+'.ckpt')
# Validation
y_pred_val, acc = test_step(W_val, val_sent_lengths, d1_val, d2_val, T_val, Y_val)
y_true_val = np.argmax(Y_val, 1)
# print 'y_true_val', np.shape(y_true_val)
# print 'y_pred_val', np.shape(y_pred_val)
fscore_val.append( f1_score(y_true_val, y_pred_val, [1,2,3,4], average='micro') )
val_res.append([y_true_val, y_pred_val])
# Testing
y_pred_test, acc = test_step(W_test, test_sent_lengths, d1_test, d2_test, T_test, Y_test)
y_true_test = np.argmax(Y_test, 1)
fscore_test.append( f1_score(y_true_test, y_pred_test, [1,2,3,4], average='micro') )
test_res.append([y_true_test, y_pred_test])
#print 'val',fscore_val
#print 'test',fscore_test
ind = np.argmax(fscore_val) #Best epoch from validation set
y_true,y_pred = test_res[ind] #actual prediction
fp.write('\n Results in Test Set (Best Index) '+str(ind)+'\n')
fp.write(str(precision_score(y_true, y_pred,[1,2,3,4], average='micro' )))
fp.write('\t')
fp.write(str(recall_score(y_true, y_pred, [1,2,3,4], average='micro' )))
fp.write('\t')
fp.write(str(f1_score(y_true, y_pred, [1,2,3,4], average='micro' )))
fp.write('\t')
fp.write('\n')
fp.write('class 1\t')
fp.write(str(precision_score(y_true, y_pred,[1], average='micro' )))
fp.write('\t')
fp.write(str(recall_score(y_true, y_pred, [1], average='micro' )))
fp.write('\t')
fp.write(str(f1_score(y_true, y_pred, [1], average='micro' )))
fp.write('\n')
fp.write('class 2\t')
fp.write(str(precision_score(y_true, y_pred,[2], average='micro' )))
fp.write('\t')
fp.write(str(recall_score(y_true, y_pred, [2], average='micro' )))
fp.write('\t')
fp.write(str(f1_score(y_true, y_pred, [2], average='micro' )))
fp.write('\n')
fp.write('class 3\t')
fp.write(str(precision_score(y_true, y_pred,[3], average='micro' )))
fp.write('\t')
fp.write(str(recall_score(y_true, y_pred, [3], average='micro' )))
fp.write('\t')
fp.write(str(f1_score(y_true, y_pred, [3], average='micro' )))
fp.write('\n')
fp.write('class 4\t')
fp.write(str(precision_score(y_true, y_pred,[4], average='micro' )))
fp.write('\t')
fp.write(str(recall_score(y_true, y_pred, [4], average='micro' )))
fp.write('\t')
fp.write(str(f1_score(y_true, y_pred, [4], average='micro' )))
fp.write('\n')
fp.write(str(confusion_matrix(y_true, y_pred)))
fp.write('\n')
for sent, slen, y_t, y_p in zip(W_test, test_sent_lengths, y_true, y_pred) :
sent_l = [str(rev_word_dict[sent[kk]]) for kk in range(slen) ]
s = ' '.join(sent_l)
fsent.write(s)
fsent.write('\n')
fsent.write( rev_label_dict[y_t] )
fsent.write('\n')
fsent.write( rev_label_dict[y_p] )
fsent.write('\n')
fsent.write('\n')
fsent.close()
rnn.sess.close()