-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathmodel.py
193 lines (137 loc) · 8.87 KB
/
model.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
import tensorflow as tf
import numpy as np
def lstm(rnn_size, keep_prob,reuse=False):
lstm_cell =tf.nn.rnn_cell.LSTMCell(rnn_size,reuse=reuse)
drop =tf.nn.rnn_cell.DropoutWrapper(lstm_cell, output_keep_prob=keep_prob)
return drop
def model_input():
input_data = tf.placeholder(tf.int32, [None, None],name='input')
target_data = tf.placeholder(tf.int32, [None, None],name='target')
input_data_len = tf.placeholder(tf.int32,[None],name='input_len')
target_data_len = tf.placeholder(tf.int32,[None],name='target_len')
lr_rate = tf.placeholder(tf.float32,name='lr')
keep_prob = tf.placeholder(tf.float32,name='keep_prob')
return input_data,target_data,input_data_len,target_data_len,lr_rate,keep_prob
def encoder_input(source_vocab_size,embed_size,input_data):
encoder_embeddings = tf.Variable(tf.random_uniform([source_vocab_size, embed_size], -1, 1))
encoder_embedded = tf.nn.embedding_lookup(encoder_embeddings, input_data)
return encoder_embedded
def encoder_layer(stacked_cells,encoder_embedded,input_data_len):
((encoder_fw_outputs,encoder_bw_outputs),
(encoder_fw_final_state,encoder_bw_final_state)) = tf.nn.bidirectional_dynamic_rnn(cell_fw=stacked_cells,
cell_bw=stacked_cells,
inputs=encoder_embedded,
sequence_length=input_data_len,
dtype=tf.float32)
encoder_outputs = tf.concat((encoder_fw_outputs,encoder_bw_outputs),2)
encoder_state_c = tf.concat((encoder_fw_final_state.c,encoder_bw_final_state.c),1)
encoder_state_h = tf.concat((encoder_fw_final_state.h,encoder_bw_final_state.h),1)
encoder_states = tf.nn.rnn_cell.LSTMStateTuple(c=encoder_state_c,h=encoder_state_h)
return encoder_outputs,encoder_states
def attention_layer(rnn_size,encoder_outputs,dec_cell,target_data_len,batch_size,encoder_states):
attention_mechanism = tf.contrib.seq2seq.BahdanauAttention(rnn_size*2,encoder_outputs,
memory_sequence_length=target_data_len)
attention_cell = tf.contrib.seq2seq.AttentionWrapper(dec_cell, attention_mechanism,
attention_layer_size=rnn_size/2)
state = attention_cell.zero_state(dtype=tf.float32, batch_size=batch_size)
state = state.clone(cell_state=encoder_states)
return attention_cell
def decoder_embedding(target_vocab_size,embed_size,decoder_input):
decoder_embeddings = tf.Variable(tf.random_uniform([target_vocab_size, embed_size], -1, 1))
dec_cell_inputs = tf.nn.embedding_lookup(decoder_embeddings, decoder_input)
return decoder_embeddings,dec_cell_inputs
def decoder_input(target_data,batch_size,vocabs_to_index):
main = tf.strided_slice(target_data, [0, 0], [batch_size, -1], [1, 1])
decoder_input = tf.concat([tf.fill([batch_size, 1],vocabs_to_index['<GO>']), main], 1)
return decoder_input
def decoder_train_layer(rnn_size,decoder_input,
dec_cell_inputs,target_vocab_size,target_data_len,
encoder_outputs,encoder_states,batch_size,attention_cell,state,dense_layer):
train_helper = tf.contrib.seq2seq.TrainingHelper(dec_cell_inputs, target_data_len)
decoder_train = tf.contrib.seq2seq.BasicDecoder(cell=attention_cell, helper=train_helper,
initial_state=state,
output_layer=dense_layer)
outputs_train, _, _ = tf.contrib.seq2seq.dynamic_decode(decoder_train,
impute_finished=True,
maximum_iterations=tf.reduce_max(target_data_len))
return outputs_train
def decoder_infer_layer(decoder_embeddings,batch_size,vocabs_to_index,
attention_cell,state,dense_layer,target_data_len):
infer_helper = tf.contrib.seq2seq.GreedyEmbeddingHelper(decoder_embeddings,
tf.fill([batch_size], vocabs_to_index['<GO>']),
vocabs_to_index['<EOS>'])
decoder_infer = tf.contrib.seq2seq.BasicDecoder(cell=attention_cell, helper=infer_helper,
initial_state=state,
output_layer=dense_layer)
outputs_infer, _, _ = tf.contrib.seq2seq.dynamic_decode(decoder_infer, impute_finished=True,
maximum_iterations=tf.reduce_max(target_data_len))
return outputs_infer
def opt_loss(outputs_train,outputs_infer,target_data_len,target_data,lr_rate):
training_logits = tf.identity(outputs_train.rnn_output, name='logits')
inference_logits = tf.identity(outputs_infer.sample_id, name='predictions')
masks = tf.sequence_mask(target_data_len, tf.reduce_max(target_data_len), dtype=tf.float32, name='masks')
cost = tf.contrib.seq2seq.sequence_loss(training_logits,target_data,masks)
optimizer = tf.train.AdamOptimizer(lr_rate)
gradients = optimizer.compute_gradients(cost)
capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in gradients if grad is not None]
train_op = optimizer.apply_gradients(capped_gradients)
return inference_logits,cost,train_op
def pad_sentence(sentence_batch, pad_int):
padded_seqs = []
seq_lens = []
max_sentence_len = max([len(sentence) for sentence in sentence_batch])
for sentence in sentence_batch:
padded_seqs.append(sentence + [pad_int] * (max_sentence_len - len(sentence)))
seq_lens.append(len(sentence))
return padded_seqs, seq_lens
def get_accuracy(target, logits):
max_seq = max(len(target[1]), logits.shape[1])
if max_seq - len(target[1]):
target = np.pad(
target,
[(0,0),(0,max_seq - len(target[1]))],
'constant')
if max_seq - logits.shape[1]:
logits = np.pad(
logits,
[(0,0),(0,max_seq - logits.shape[1])],
'constant')
return np.mean(np.equal(target, logits))
def sentence_to_seq(sentence, vocabs_to_index):
results = []
for word in sentence.split(" "):
if word in vocabs_to_index:
results.append(vocabs_to_index[word])
else:
results.append(vocabs_to_index['<UNK>'])
return results
def decoder_layer(rnn_size,encoder_outputs,target_data_len,
dec_cell,encoder_states,target_data,vocabs_to_index,target_vocab_size,
embed_size,dense_layer,attention_cell,state,batch_size):
decoder_input_tensor = decoder_input(target_data,batch_size,vocabs_to_index)
decoder_embeddings,dec_cell_inputs = decoder_embedding(target_vocab_size,embed_size,decoder_input_tensor)
outputs_train = decoder_train_layer(rnn_size,decoder_input_tensor,dec_cell_inputs,target_vocab_size,target_data_len,encoder_outputs,encoder_states,batch_size,attention_cell,state,dense_layer)
outputs_infer = decoder_infer_layer(decoder_embeddings,batch_size,vocabs_to_index,attention_cell,state,dense_layer,target_data_len)
return outputs_train,outputs_infer
def seq2seq_model(source_vocab_size,embed_size,rnn_size,keep_prob,
target_vocab_size,batch_size,vocabs_to_index):
input_data,target_data,input_data_len,target_data_len,lr_rate,keep_probs = model_input()
encoder_embedded = encoder_input(source_vocab_size,embed_size,input_data)
stacked_cells = lstm(rnn_size, keep_prob)
encoder_outputs,encoder_states = encoder_layer(stacked_cells,
encoder_embedded,
input_data_len)
dec_cell = lstm(rnn_size*2,keep_prob)
dense_layer = tf.layers.Dense(target_vocab_size)
attention_mechanism = tf.contrib.seq2seq.BahdanauAttention(rnn_size*2,encoder_outputs,
memory_sequence_length=target_data_len)
attention_cell = tf.contrib.seq2seq.AttentionWrapper(dec_cell, attention_mechanism,
attention_layer_size=rnn_size/2)
state = attention_cell.zero_state(dtype=tf.float32, batch_size=batch_size)
state = state.clone(cell_state=encoder_states)
# attention_cell = attention_layer(rnn_size,encoder_outputs,target_data_len,dec_cell,batch_size,encoder_states)
outputs_train,outputs_infer = decoder_layer(rnn_size,encoder_outputs,target_data_len,
dec_cell,encoder_states,target_data,vocabs_to_index,target_vocab_size,
embed_size,dense_layer,attention_cell,state,batch_size)
inference_logits,cost,train_op = opt_loss(outputs_train,outputs_infer,target_data_len,target_data,lr_rate)
return input_data,target_data,input_data_len,target_data_len,lr_rate,keep_probs,inference_logits,cost,train_op