-
Notifications
You must be signed in to change notification settings - Fork 35
/
Copy pathlstm.py
177 lines (134 loc) · 6.4 KB
/
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
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
# -*- coding: utf-8 -*-
"""
Created on Tue Apr 3 09:31:45 2018
@author: moseli
"""
from numpy.random import seed
seed(1)
from sklearn.model_selection import train_test_split as tts
import logging
import plotly.plotly as py
import plotly.graph_objs as go
import matplotlib.pyplot as plt
import pandas as pd
import pydot
import keras
from keras import backend as k
k.set_learning_phase(1)
from keras.preprocessing.text import Tokenizer
from keras import initializers
from keras.optimizers import RMSprop
from keras.models import Sequential,Model
from keras.layers import Dense,LSTM,Dropout,Input,Activation,Add,Concatenate
from keras.layers.advanced_activations import LeakyReLU,PReLU
from keras.callbacks import ModelCheckpoint
from keras.models import load_model
from keras.optimizers import Adam
#keras.utils.vis_utils import plot_model
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s',\
level=logging.INFO)
#######################model params###########################
batch_size = 100
num_classes = 1
epochs = 5
hidden_units = emb_size_all
learning_rate = 0.002
clip_norm = 1.0
###############################################################
en_shape=np.shape(train_data["article"][0])
de_shape=np.shape(train_data["summaries"][0])
"""______generate summary length for test______"""
#train_data["nums_summ"]=list(map(lambda x:0 if len(x)<5000 else 1,data["articles"]))
#train_data["nums_summ"]=list(map(len,data["summaries"]))
#train_data["nums_summ_norm"]=(np.array(train_data["nums_summ"])-min(train_data["nums_summ"]))/(max(train_data["nums_summ"])-min(train_data["nums_summ"]))
############################helpers###########################
def encoder_decoder(data):
print('Encoder_Decoder LSTM...')
"""__encoder___"""
encoder_inputs = Input(shape=en_shape)
encoder_LSTM = LSTM(hidden_units, dropout_U = 0.2, dropout_W = 0.2 ,return_state=True)
encoder_LSTM_rev=LSTM(hidden_units,return_state=True,go_backwards=True)
#merger=Add()[encoder_LSTM(encoder_inputs), encoder_LSTM_rev(encoder_inputs)]
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)
#plot_model(model, to_file=modelLocation+'model.png', show_shapes=True)
rmsprop = RMSprop(lr=learning_rate,clipnorm=clip_norm)
model.compile(loss='mse',optimizer=rmsprop)
x_train,x_test,y_train,y_test=tts(data["article"],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)
#plot_model(encoder_model_inf, to_file='encoder_model.png', show_shapes=True)
#plot_model(decoder_model_inf, to_file='decoder_model.png', show_shapes=True)
scores = model.evaluate([x_test,y_test],y_test, verbose=0)
print('LSTM test scores:', scores)
#announcedone()
print('\007')
print(model.summary())
return model,encoder_model_inf,decoder_model_inf
"""___pred____"""
def comparePred(index):
pred=trained_model.predict([np.reshape(train_data["article"][index],(1,en_shape[0],emb_size_all)),np.reshape(train_data["summaries"][index],(1,de_shape[0],emb_size_all))])
return pred
"""____generate summary from vectors and remove padding words___"""
def generateText(SentOfVecs):
SentOfVecs=np.reshape(SentOfVecs,de_shape)
kk=""
for k in SentOfVecs:
kk=kk+((getWord(k)[0]+" ") if getWord(k)[1]>0.2 else "")
return kk
"""___generate summary vectors___"""
def summarize(article):
stop_pred = False
article = np.reshape(article,(1,en_shape[0],en_shape[1]))
#get initial h and c values from encoder
init_state_val = encoder.predict(article)
target_seq = np.zeros((1,1,emb_size_all))
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]
#get most similar word and put in line to be input in next timestep
#target_seq=np.reshape(model.wv[getWord(decoder_out)[0]],(1,1,emb_size_all))
target_seq=np.reshape(decoder_out,(1,1,emb_size_all))
if len(generated_summary)== de_shape[0]:
stop_pred=True
break
return generated_summary
#######################################################################################
trained_model,encoder,decoder = encoder_decoder(train_data)
def saveModels():
trained_model.save("%sinit_model"%modelLocation)
encoder.save("%sencoder"%modelLocation)
decoder.save("%sdecoder"%modelLocation)
print(generateText(summarize(train_data["article"][10])))
print(data["summaries"][10])
print(data["articles"][10])
del trained_model,encoder,decoder
getWord(collect_pred[23])
model.wv.most_similar(np.zeros((1,emb_size_all)))