-
Notifications
You must be signed in to change notification settings - Fork 0
/
anwser_extracter_from_2_sentence.py
executable file
·160 lines (129 loc) · 6.62 KB
/
anwser_extracter_from_2_sentence.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
# -*- coding: utf-8 -*-
"""anwser_extracter_from_2_sentence.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1kafO9dqV-WCX0dfjTN6xGRDUulz7V6Z-
"""
!wget https://s3.amazonaws.com/text-datasets/babi_tasks_1-20_v1-2.tar.gz
!tar -xzvf babi_tasks_1-20_v1-2.tar.gz
import numpy as np
import keras
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences
from keras.utils.np_utils import to_categorical
import matplotlib.pyplot as plt
max_vocab=20000
emb_dim=15
max_sen=2
train_set=open("./tasks_1-20_v1-2/en/qa1_single-supporting-fact_train.txt")
test_set=open("./tasks_1-20_v1-2/en/qa1_single-supporting-fact_test.txt")
def getting_x_y(data):
first_sen_story=[]
second_sen_story=[]
questions=[]
answers=[]
t=1
for line in data:
line=line[2:]
if t==1:
first_sen_story.append(line)
t+=1
elif t==2:
second_sen_story.append(line)
t+=1
else:
q,a,_=line.split("\t")
questions.append(q)
answers.append(a)
t=1
return first_sen_story,second_sen_story,questions,answers
train_first_sen_story,train_second_sen_story,train_questions,train_answers=getting_x_y(train_set)
test_first_sen_story,test_second_sen_story,test_questions,test_answers=getting_x_y(test_set)
story_tokenizer=Tokenizer(num_words=max_vocab)
story_tokenizer.fit_on_texts(train_first_sen_story+train_second_sen_story+test_first_sen_story+test_second_sen_story)
train_first_seq_story=story_tokenizer.texts_to_sequences(train_first_sen_story)
train_second_seq_story=story_tokenizer.texts_to_sequences(train_second_sen_story)
test_first_seq_story=story_tokenizer.texts_to_sequences(test_first_sen_story)
test_second_seq_story=story_tokenizer.texts_to_sequences(test_second_sen_story)
max_story_sen_length=max(max(len(s) for s in train_first_seq_story ),max(len(s) for s in train_second_seq_story ),max(len(s) for s in test_first_seq_story ),max(len(s) for s in test_second_seq_story ))
train_first_seq_story=pad_sequences(train_first_seq_story,maxlen =max_story_sen_length,padding = "post")
train_second_seq_story=pad_sequences(train_second_seq_story,maxlen =max_story_sen_length,padding = "post")
test_first_seq_story=pad_sequences(test_first_seq_story,maxlen =max_story_sen_length,padding = "post")
test_second_seq_story=pad_sequences(test_second_seq_story,maxlen =max_story_sen_length,padding = "post")
questions_tokenizer=Tokenizer(num_words=max_vocab)
questions_tokenizer.fit_on_texts(train_questions+test_questions)
train_seq_questions=questions_tokenizer.texts_to_sequences(train_questions)
test_seq_questions=questions_tokenizer.texts_to_sequences(test_questions)
max_questions_sen_length=max(max(len(s) for s in train_seq_questions ),max(len(s) for s in test_seq_questions) )
train_seq_questions=pad_sequences(train_seq_questions,maxlen =max_questions_sen_length,padding = "post")
test_seq_questions=pad_sequences(test_seq_questions,maxlen =max_questions_sen_length,padding = "post")
answers_tokenizer=Tokenizer(num_words=max_vocab)
answers_tokenizer.fit_on_texts(train_answers+test_answers)
train_seq_answers=answers_tokenizer.texts_to_sequences(train_answers)
test_seq_answers=answers_tokenizer.texts_to_sequences(test_answers)
max_answers_sen_length=max(max(len(s) for s in train_seq_answers ),max(len(s) for s in test_seq_answers) )
train_seq_answers=pad_sequences(train_seq_answers,maxlen =max_answers_sen_length,padding = "post")
test_seq_answers=pad_sequences(test_seq_answers,maxlen =max_answers_sen_length,padding = "post")
def stack_stories(data1,data2):
data=[]
for i in range(len(data1)):
data.append([data1[i],data2[i]])
return np.asarray(data)
train_stories=stack_stories(train_first_seq_story,train_second_seq_story)
test_stories=stack_stories(test_first_seq_story,test_second_seq_story)
word2index=answers_tokenizer.word_index
train_seq_answers=to_categorical(train_seq_answers)
test_seq_answers=to_categorical(test_seq_answers)
story_input=keras.layers.Input(shape=(max_sen,max_story_sen_length,))
story_emb=keras.layers.Embedding(max_vocab,emb_dim)
x=story_emb(story_input)
# the output of x is (2,max_story_sen_length,emb_dim)
# model=keras.Model(story_input,x)
sum_of_two_stories=keras.layers.Lambda(lambda x:keras.backend.sum(x,axis=2))
s=sum_of_two_stories(x)
# the output of s is (1,max_story_sen_length,emb_dim)
# so its only add them
# model1=keras.Model(story_input,s)
question_input=keras.layers.Input(shape=(max_questions_sen_length,))
question_emb=keras.layers.Embedding(max_vocab,emb_dim)
output_of_question_emb=question_emb(question_input)
sum_question=keras.layers.Lambda(lambda x:keras.backend.sum(x,axis=1))
output_of_sum_question=sum_question(output_of_question_emb)
# the output of output_of_question_emb is (1,emb_dim)
# so its only add them
question_reshape=keras.layers.Reshape((1,emb_dim))
output_of_question_reshape=question_reshape(output_of_sum_question)
# it reshape them into 3 dim only
# model2=keras.Model(question_input,output_of_sum_question)
# model3=keras.Model(question_input,output_of_question_reshape)
mul_story_with_question=keras.layers.dot([s,output_of_question_reshape],2)
flatten_=keras.layers.Flatten()
the_output_flatten=flatten_(mul_story_with_question)
# model4=keras.Model([story_input,question_input],mul_story_with_question)
# model5=keras.Model([story_input,question_input],the_output_flatten)
get_story_weights=keras.layers.Activation('softmax')(the_output_flatten)
story_weigths=keras.layers.Reshape((max_sen,1))(get_story_weights)
debug_model=keras.Model([story_input,question_input],story_weigths)
mul_weight_with_story=keras.layers.dot([story_weigths,s],1)
flatten2_=keras.layers.Flatten()
y=flatten2_(mul_weight_with_story)
y=keras.layers.Dense(len(word2index)+1,activation="softmax")(y)
model=keras.Model([story_input,question_input],y)
model.compile("adam",loss=keras.losses.CategoricalCrossentropy(),metrics=["accuracy"])
history=model.fit([train_stories,train_seq_questions],train_seq_answers,batch_size=32,epochs=50
,validation_data=([test_stories,test_seq_questions],test_seq_answers))
plt.plot(history.history["accuracy"],label="accuracy")
plt.plot(history.history["val_accuracy"],label="val_accuracy")
plt.legend()
plt.show()
plt.plot(history.history["loss"],label="loss")
plt.plot(history.history["val_loss"],label="loss")
plt.legend()
plt.show()
i=np.random.choice(len(train_first_sen_story))
z=model.predict([train_stories[i:i+1],train_seq_questions[i:i+1]])
idx=np.argmax( z[0,:])
print(train_first_sen_story[i])
print(train_second_sen_story[i])
print(train_questions[i])
answers_tokenizer.index_word[idx]