调用外部的函数介绍
tf.sampled_softmax_loss()
tf.sampled_softmax_loss()中调用了_compute_sampled_logits() 关于__compute_sampled_logits()
#此函数和nce_loss是差不多的, 取样求loss
def sampled_softmax_loss(weights, #[num_classes, dim]
biases, #[num_classes]
inputs, #[batch_size, dim]
labels, #[batch_size, num_true]
num_sampled,
num_classes,
num_true=1,
sampled_values=None,
remove_accidental_hits=True,
partition_strategy="mod",
name="sampled_softmax_loss"):
#return: [batch_size]
关于参数labels:一般情况下,num_true为1, labels的shpae为[batch_size, 1]。假设我们有1000个类别, 使用one_hot形式的label的话, 我们的labels的shape是[batch_size, num_classes]。显然,如果num_classes非常大的话,会影响计算性能。所以,这里采用了一个稀疏的方式,即:使用3代表了[0,0,0,1,0….]
tf.nn.seq2seq.embedding_attention_seq2seq()
创建了input embedding matrix 和 output embedding matrix
def embedding_attention_seq2seq(encoder_inputs, #[T, batch_size]
decoder_inputs, #[out_T, batch_size]
cell,
num_encoder_symbols,
num_decoder_symbols,
embedding_size,
num_heads=1, #只采用一个read head
output_projection=None,
feed_previous=False,
dtype=None,
scope=None,
initial_state_attention=False):
#output_projection: (W, B) W:[output_size, num_decoder_symbols]
#B: [num_decoder_symbols]
(1)这个函数创建了一个inputs 的 embedding matrix.
(2)计算了encoder的 output,并保存起来,用于计算attention
encoder_cell = rnn_cell.EmbeddingWrapper(
cell, embedding_classes=num_encoder_symbols,
embedding_size=embedding_size)# 创建了inputs的 embedding matrix
encoder_outputs, encoder_state = rnn.rnn(
encoder_cell, encoder_inputs, dtype=dtype) #return [T ,batch_size,size]
(3)生成attention states
top_states = [array_ops.reshape(e, [-1, 1, cell.output_size])
for e in encoder_outputs] # T * batch_size * 1 * size
attention_states = array_ops.concat(1, top_states) # batch_size*T*size
(4)剩下的工作交给embedding_attention_decoder,embedding_attention_decoder中创建了decoder的embedding matrix
# Decoder.
output_size = None
if output_projection is None:
cell = rnn_cell.OutputProjectionWrapper(cell, num_decoder_symbols)
output_size = num_decoder_symbols
if isinstance(feed_previous, bool):
return embedding_attention_decoder(
decoder_inputs,
encoder_state,
attention_states,
cell,
num_decoder_symbols,
embedding_size,
num_heads=num_heads,
output_size=output_size,
output_projection=output_projection,
feed_previous=feed_previous,
initial_state_attention=initial_state_attention)
tf.nn.rnn_cell.EmbeddingWrapper()
embedding_attention_seq2seq中调用了这个类
使用了这个类之后,rnn 的inputs就可以是[batch_size]了,里面保存的是word的id。
此类就是在 cell 前 加了一层embedding
class EmbeddingWrapper(RNNCell):
def __init__(self, cell, embedding_classes, embedding_size, initializer=None):
def __call__(self, inputs, state, scope=None):#生成embedding矩阵[embedding_classes,embedding_size]
#inputs: [batch_size, 1]
#return : (output, state)
tf.nn.rnn_cell.OutputProgectionWrapper()
将rnn_cell的输出映射成想要的维度
class OutputProjectionWrapper(RNNCell):
def __init__(self, cell, output_size): # output_size:映射后的size
def __call__(self, inputs, state, scope=None):
#init 返回一个带output projection的 rnn_cell
tf.nn.seq2seq.embedding_attention_decoder()
#生成embedding matrix :[num_symbols, embedding_size]
def embedding_attention_decoder(decoder_inputs, # T*batch_size
initial_state,
attention_states,
cell,
num_symbols,
embedding_size,
num_heads=1,
output_size=None,
output_projection=None,
feed_previous=False,
update_embedding_for_previous=True,
dtype=None,
scope=None,
initial_state_attention=False):
#核心代码
embedding = variable_scope.get_variable("embedding",
[num_symbols, embedding_size]) #output embedding
loop_function = _extract_argmax_and_embed(
embedding, output_projection,
update_embedding_for_previous) if feed_previous else None
emb_inp = [
embedding_ops.embedding_lookup(embedding, i) for i in decoder_inputs]
return attention_decoder(
emb_inp,
initial_state,
attention_states,
cell,
output_size=output_size,
num_heads=num_heads,
loop_function=loop_function,
initial_state_attention=initial_state_attention)
可以看到,此函数先为 decoder symbols 创建了一个embedding矩阵。然后定义了loop_function。
emb_in是embedded input :[T, batch_size, embedding_size]
函数的主要工作还是交给了attention_decoder()
tf.nn.attention_decoder()
def attention_decoder(decoder_inputs, #[T, batch_size, input_size]
initial_state, #[batch_size, cell.states]
attention_states, #[batch_size , attn_length , attn_size]
cell,
output_size=None,
num_heads=1,
loop_function=None,
dtype=None,
scope=None,
initial_state_attention=False):
论文中,在计算attention distribution的时候,提到了三个公式
其中, W1 W 1 维度是[attn_vec_size, size], hi h i :[size,1], 我们日常表示输入数据,都是用列向量表示,但是在tensorflow中,趋向用行向量表示。在这个函数中,为了计算 W1∗hi W 1 ∗ h i , 使用了卷积的方式。
hidden = array_ops.reshape(
attention_states, [-1, attn_length, 1, attn_size]) #[batch_size * T * 1 * input_size]
hidden_features = []
v = []
attention_vec_size = attn_size # Size of query vectors for attention.
for a in xrange(num_heads):
k = variable_scope.get_variable("AttnW_%d" % a,
[1, 1, attn_size, attention_vec_size])
hidden_features.append(nn_ops.conv2d(hidden, k, [1, 1, 1, 1], "SAME"))
v.append(
variable_scope.get_variable("AttnV_%d" % a, [attention_vec_size])) #attention_vec_size = attn_size
使用conv2d之后,返回的tensor的形状是[batch_size, attn_length, 1, attention_vec_size]
此函数是这么求
W2∗dt
W
2
∗
d
t
和
si
s
i
的。
y = linear(query, attention_vec_size, True)
y = array_ops.reshape(y, [-1, 1, 1, attention_vec_size])
# Attention mask is a softmax of v^T * tanh(...).
s = math_ops.reduce_sum(
v[a] * math_ops.tanh(hidden_features[a] + y), [2, 3]) #[batch_size, attn_length, 1, attn_size]
a = nn_ops.softmax(s) #s" [batch_size * attn_len]
# Now calculate the attention-weighted vector d.
d = math_ops.reduce_sum(
array_ops.reshape(a, [-1, attn_length, 1, 1]) * hidden,
[1, 2])
ds.append(array_ops.reshape(d, [-1, attn_size]))
y=W2∗dt,d=d′ y = W 2 ∗ d t , d = d ′
def rnn()
from tensorflow.python.ops import rnn
rnn.rnn()
def rnn(cell, inputs, initial_state=None, dtype=None,
sequence_length=None, scope=None):
#inputs: A length T list of inputs, each a `Tensor` of shape`[batch_size, input_size]`
#sequence_length: [batch_size], 指定sample 序列的长度
#return : (outputs, states), outputs: T*batch_size*output_size. states:batch_size*state
seq2seqModel
- 创建映射参数 proj_w, proj_b
- 声明:sampled_loss,看了word2vec的就会理解
- 声明:seq2seq_f(),构建了inputs的embedding和outputs的embedding,进行核心计算
- 使用model_with_buckets(),model_with_buckets中调用了seq2seq_f和 sampled_loss
model_with_buckets()
调用方法 tf.nn.seq2seq.model_with_buckets()
def model_with_buckets(encoder_inputs, decoder_inputs, targets, weights,
buckets, seq2seq, softmax_loss_function=None,
per_example_loss=False, name=None):
"""Create a sequence-to-sequence model with support for bucketing.
The seq2seq argument is a function that defines a sequence-to-sequence model,
e.g., seq2seq = lambda x, y: basic_rnn_seq2seq(x, y, rnn_cell.GRUCell(24))
Args:
encoder_inputs: A list of Tensors to feed the encoder; first seq2seq input.
decoder_inputs: A list of Tensors to feed the decoder; second seq2seq input.
targets: A list of 1D batch-sized int32 Tensors (desired output sequence).
weights: List of 1D batch-sized float-Tensors to weight the targets.
buckets: A list of pairs of (input size, output size) for each bucket.
seq2seq: A sequence-to-sequence model function; it takes 2 input that agree with encoder_inputs and decoder_inputs, and returns a pair consisting of outputs and states (as, e.g., basic_rnn_seq2seq).
softmax_loss_function: Function (inputs-batch, labels-batch) -> loss-batch
to be used instead of the standard softmax (the default if this is None).
per_example_loss: Boolean. If set, the returned loss will be a batch-sized
tensor of losses for each sequence in the batch. If unset, it will be
a scalar with the averaged loss from all examples.
name: Optional name for this operation, defaults to "model_with_buckets".
Returns:
A tuple of the form (outputs, losses), where:
outputs: The outputs for each bucket. Its j'th element consists of a list
of 2D Tensors. The shape of output tensors can be either
[batch_size x output_size] or [batch_size x num_decoder_symbols]
depending on the seq2seq model used.
losses: List of scalar Tensors, representing losses for each bucket, or,
if per_example_loss is set, a list of 1D batch-sized float Tensors.
Raises:
ValueError: If length of encoder_inputsut, targets, or weights is smaller
than the largest (last) bucket.
"""
记住,tensorflow的编码方法是:先构图,再训练。训练是根据feed确定的