-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbirectional_lstm.py
137 lines (104 loc) · 5.05 KB
/
birectional_lstm.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
# -*- coding: utf-8 -*-
"""
Created on Sat Apr 13 15:04:19 2019
@author: sumedh
"""
import os
import load_data as ld
import w2v_features as w2v
import numpy as np
from sklearn.model_selection import train_test_split as tts
from keras import backend as K
from keras.layers import Dense, LSTM, Input, Add, TimeDistributed, Flatten, RepeatVector, merge, Lambda, Multiply, Reshape, Permute
from keras.callbacks import ModelCheckpoint
from keras.models import Model, load_model
from keras.optimizers import RMSprop
def encoder_decoder(data, en_shape, de_shape, hidden_units, learning_rate, clip_norm, epochs, batch_size):
''' encoder '''
encoder_inputs = Input(shape=en_shape)
encoder_LSTM = LSTM(hidden_units, dropout = 0.02 ,return_state=True)
encoder_LSTM_rev=LSTM(hidden_units,return_state=True,go_backwards=True)
encoder_outputsR, state_hR, state_cR = encoder_LSTM_rev(encoder_inputs)
encoder_outputs, state_h, state_c = encoder_LSTM(encoder_inputs)
state_hfinal=Add()([state_h,state_hR])
state_cfinal=Add()([state_c,state_cR])
encoder_states = [state_hfinal,state_cfinal]
''' decoder '''
decoder_inputs = Input(shape=(None,de_shape[1]))
decoder_LSTM = LSTM(hidden_units,return_sequences=True,return_state=True)
decoder_outputs, _, _ = decoder_LSTM(decoder_inputs,initial_state=encoder_states)
decoder_dense = Dense(de_shape[1],activation='linear')
decoder_outputs = decoder_dense(decoder_outputs)
model= Model(inputs=[encoder_inputs,decoder_inputs], outputs=decoder_outputs)
rmsprop = RMSprop(lr=learning_rate,clipnorm=clip_norm)
model.compile(loss='mse',optimizer=rmsprop)
x_train,x_test,y_train,y_test=tts(data["review"],data["summaries"],test_size=0.20)
model.fit(x=[x_train,y_train],
y=y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=([x_test,y_test], y_test))
''' inference mode '''
encoder_model_inf = Model(encoder_inputs,encoder_states)
decoder_state_input_H = Input(shape=(hidden_units,))
decoder_state_input_C = Input(shape=(hidden_units,))
decoder_state_inputs = [decoder_state_input_H, decoder_state_input_C]
decoder_outputs, decoder_state_h, decoder_state_c = decoder_LSTM(decoder_inputs,
initial_state=decoder_state_inputs)
decoder_states = [decoder_state_h, decoder_state_c]
decoder_outputs = decoder_dense(decoder_outputs)
decoder_model_inf= Model([decoder_inputs]+decoder_state_inputs,
[decoder_outputs]+decoder_states)
#scores = model.evaluate([x_test,y_test],y_test, verbose=0)
return model,encoder_model_inf,decoder_model_inf
def summarize(review, en_shape, encoder, decoder,max_len = 300):
stop_pred = False
review = np.reshape(review, (1, en_shape[0], en_shape[1]))
init_state_val = encoder.predict(review)
target_seq = np.zeros((1,1,300))
generated_summary=[]
while not stop_pred:
decoder_out,decoder_h,decoder_c= decoder.predict(x=[target_seq]+init_state_val)
generated_summary.append(decoder_out)
init_state_val= [decoder_h,decoder_c]
target_seq=np.reshape(decoder_out,(1,1,300))
if len(generated_summary)== max_len:
stop_pred=True
break
#sent_vecs = np.reshape(generated_summary, de_shape)
summ = ''
for k in generated_summary:
try:
summ = summ + ohe.inverse_transform([np.argmax(k)])[0].strip()+" "
except:
summ = summ +' UNK '
return summ
path = os.getcwd()
reviews = path+'\\summaries handwritten\\'
datasets={'reviews':reviews}
data_categories=["training","validation","test"]
filenames= ld.load_data(datasets['reviews'],data_categories[0])
data = ld.make_data_dict(filenames,datasets,data_categories)
corp = w2v.create_corpus(data)
embed_model = w2v.word2vec_model(corpus = corp, loc = path+'\\train_w2v_embedding\\')
ohe, onehot_encoded,onehot = w2v.one_hot_encode(corpus = corp, vocab = embed_model.wv.index2word)
train_data = w2v.w2v_matrix(embed_model,data)
train_data= w2v.cut_seq(train_data,1,5)
train_data["summaries"]=np.array(train_data["summaries"])
train_data["review"]=np.array(train_data["review"])
train_data["summaries"]=np.array(list(map(w2v.addones,train_data["summaries"])))
trained_model,encoder,decoder = encoder_decoder(data = train_data,
en_shape = train_data['review'][0].shape,
de_shape = train_data['summaries'][0].shape,
hidden_units = 1500,
learning_rate = 0.05,
clip_norm = 2,
epochs = 20,
batch_size = 30)
print(summarize(review = train_data["review"][1],
en_shape = train_data['review'][1].shape,
# de_shape = train_data['summaries'][5].shape,
max_len = 10,
encoder = encoder,
decoder = decoder))