-
Notifications
You must be signed in to change notification settings - Fork 100
/
Copy pathseq2seq_model.py
127 lines (118 loc) · 7.02 KB
/
seq2seq_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
import tensorflow as tf
from seq2seq import embedding_attention_seq2seq
class Seq2SeqModel():
def __init__(self, source_vocab_size, target_vocab_size, en_de_seq_len, hidden_size, num_layers,
batch_size, learning_rate, num_samples=1024,
forward_only=False, beam_search=True, beam_size=10):
'''
初始化并创建模型
:param source_vocab_size:encoder输入的vocab size
:param target_vocab_size: decoder输入的vocab size，这里跟上面一样
:param en_de_seq_len: 源和目的序列最大长度
:param hidden_size: RNN模型的隐藏层单元个数
:param num_layers: RNN堆叠的层数
:param batch_size: batch大小
:param learning_rate: 学习率
:param num_samples: 计算loss时做sampled softmax时的采样数
:param forward_only: 预测时指定为真
:param beam_search: 预测时是采用greedy search还是beam search
:param beam_size: beam search的大小
'''
self.source_vocab_size = source_vocab_size
self.target_vocab_size = target_vocab_size
self.en_de_seq_len = en_de_seq_len
self.hidden_size = hidden_size
self.num_layers = num_layers
self.batch_size = batch_size
self.learning_rate = tf.Variable(float(learning_rate), trainable=False)
self.num_samples = num_samples
self.forward_only = forward_only
self.beam_search = beam_search
self.beam_size = beam_size
self.global_step = tf.Variable(0, trainable=False)
output_projection = None
softmax_loss_function = None
# 定义采样loss函数，传入后面的sequence_loss_by_example函数
if num_samples > 0 and num_samples < self.target_vocab_size:
w = tf.get_variable('proj_w', [hidden_size, self.target_vocab_size])
w_t = tf.transpose(w)
b = tf.get_variable('proj_b', [self.target_vocab_size])
output_projection = (w, b)
#调用sampled_softmax_loss函数计算sample loss，这样可以节省计算时间
def sample_loss(logits, labels):
labels = tf.reshape(labels, [-1, 1])
return tf.nn.sampled_softmax_loss(w_t, b, labels=labels, inputs=logits, num_sampled=num_samples, num_classes=self.target_vocab_size)
softmax_loss_function = sample_loss
self.keep_drop = tf.placeholder(tf.float32)
# 定义encoder和decoder阶段的多层dropout RNNCell
def create_rnn_cell():
encoDecoCell = tf.contrib.rnn.BasicLSTMCell(hidden_size)
encoDecoCell = tf.contrib.rnn.DropoutWrapper(encoDecoCell, input_keep_prob=1.0, output_keep_prob=self.keep_drop)
return encoDecoCell
encoCell = tf.contrib.rnn.MultiRNNCell([create_rnn_cell() for _ in range(num_layers)])
# 定义输入的placeholder，采用了列表的形式
self.encoder_inputs = []
self.decoder_inputs = []
self.decoder_targets = []
self.target_weights = []
for i in range(en_de_seq_len[0]):
self.encoder_inputs.append(tf.placeholder(tf.int32, shape=[None, ], name="encoder{0}".format(i)))
for i in range(en_de_seq_len[1]):
self.decoder_inputs.append(tf.placeholder(tf.int32, shape=[None, ], name="decoder{0}".format(i)))
self.decoder_targets.append(tf.placeholder(tf.int32, shape=[None, ], name="target{0}".format(i)))
self.target_weights.append(tf.placeholder(tf.float32, shape=[None, ], name="weight{0}".format(i)))
# test模式，将上一时刻输出当做下一时刻输入传入
if forward_only:
if beam_search:#如果是beam_search的话，则调用自己写的embedding_attention_seq2seq函数，而不是legacy_seq2seq下面的
self.beam_outputs, _, self.beam_path, self.beam_symbol = embedding_attention_seq2seq(
self.encoder_inputs, self.decoder_inputs, encoCell, num_encoder_symbols=source_vocab_size,
num_decoder_symbols=target_vocab_size, embedding_size=hidden_size,
output_projection=output_projection, feed_previous=True)
else:
decoder_outputs, _ = tf.contrib.legacy_seq2seq.embedding_attention_seq2seq(
self.encoder_inputs, self.decoder_inputs, encoCell, num_encoder_symbols=source_vocab_size,
num_decoder_symbols=target_vocab_size, embedding_size=hidden_size,
output_projection=output_projection, feed_previous=True)
# 因为seq2seq模型中未指定output_projection，所以需要在输出之后自己进行output_projection
if output_projection is not None:
self.outputs = tf.matmul(decoder_outputs, output_projection[0]) + output_projection[1]
else:
# 因为不需要将output作为下一时刻的输入，所以不用output_projection
decoder_outputs, _ = tf.contrib.legacy_seq2seq.embedding_attention_seq2seq(
self.encoder_inputs, self.decoder_inputs, encoCell, num_encoder_symbols=source_vocab_size,
num_decoder_symbols=target_vocab_size, embedding_size=hidden_size, output_projection=output_projection,
feed_previous=False)
self.loss = tf.contrib.legacy_seq2seq.sequence_loss(
decoder_outputs, self.decoder_targets, self.target_weights, softmax_loss_function=softmax_loss_function)
# Initialize the optimizer
opt = tf.train.AdamOptimizer(learning_rate=self.learning_rate, beta1=0.9, beta2=0.999, epsilon=1e-08)
self.optOp = opt.minimize(self.loss)
self.saver = tf.train.Saver(tf.all_variables())
def step(self, session, encoder_inputs, decoder_inputs, decoder_targets, target_weights, go_token_id):
#传入一个batch的数据，并训练性对应的模型
# 构建sess.run时的feed_inpits
feed_dict = {}
if not self.forward_only:
feed_dict[self.keep_drop] = 0.5
for i in range(self.en_de_seq_len[0]):
feed_dict[self.encoder_inputs[i].name] = encoder_inputs[i]
for i in range(self.en_de_seq_len[1]):
feed_dict[self.decoder_inputs[i].name] = decoder_inputs[i]
feed_dict[self.decoder_targets[i].name] = decoder_targets[i]
feed_dict[self.target_weights[i].name] = target_weights[i]
run_ops = [self.optOp, self.loss]
else:
feed_dict[self.keep_drop] = 1.0
for i in range(self.en_de_seq_len[0]):
feed_dict[self.encoder_inputs[i].name] = encoder_inputs[i]
feed_dict[self.decoder_inputs[0].name] = [go_token_id]
if self.beam_search:
run_ops = [self.beam_path, self.beam_symbol]
else:
run_ops = [self.outputs]
outputs = session.run(run_ops, feed_dict)
if not self.forward_only:
return None, outputs[1]
else:
if self.beam_search:
return outputs[0], outputs[1]