【NLP】⚠️学不会打我! 半小时学会基本操作 10⚠️ Seq2seq
概述
从今天开始我们将开启一段自然语言处理 (NLP) 的旅程. 自然语言处理可以让来处理, 理解, 以及运用人类的语言, 实现机器语言和人类语言之间的沟通桥梁.
Seq2seq
Seq2seq 由 Encoder 和 Decoder 两个 RNN 组成. Encoder 将变长序列输出, 编码成 encoderstate 再由 Decoder 输出变长序列.
Seq2seq 的使用领域:
- 机器翻译: Encoder-Decoder 的最经典应用
- 文本摘要: 输入是一段文本序列, 输出是这段文本序列的摘要序列
- 阅读理解: 将输入的文章和问题分别编码, 再对其进行解码得到问题的答案
- 语音识别: 输入是语音信号序列, 输出是文字序列
优点:
- 非常灵活: 并不限制 Encoder, Decoder 使用何种神经网络, 也不限制输入和输出的状态
- 端到端: 将语义理解和语言生成结合在了一起, 而不是分开处理
缺点:
- 信息损失: 无论输入如何变化, Encoder 给出的都是一个固定维数的向量. 在生成文本时, 生成每个词所用到的语义向量都是一样的, 过于简单
Attention 模型
Attention 是一种用于提升 RNN 的 Encoder 和 Decoder 模型的效果的机制. 广泛应用于机器翻译, 语音识别, 图像标注等多个领域. 深度学习中的注意力机制从本质上讲和人类的选择性视觉注意力机制类似. 核心目标也是从众多信息中选择出对当前任务目标更关键的信息.
Attention 实质上是一种 content-based addressing 的机制. 即从网络中某些状态集合中选取给定状态较为相似的状态, 进而做后续的信息抽取.
首先根据 Encoder 和 Decoder 的特征计算权值, 然后对 Encoder 的特征进行加权求和, 作为 Decoder 的输入. 其作用的将 Encoder 的特征以更好的方式呈献给 Decoder. (并不是所有的 context 都对下一个状态的生成产生影响, Attention 就是选择恰当的 context 用它生成下一个状态.
Seq2seq 模型
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Sequence-to-sequence model with an attention mechanism."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import random
import numpy as np
from six.moves import xrange # pylint: disable=redefined-builtin
import tensorflow as tf
import data_utils
setattr(tf.contrib.rnn.GRUCell, '__deepcopy__', lambda self, _: self)
setattr(tf.contrib.rnn.BasicLSTMCell, '__deepcopy__', lambda self, _: self)
setattr(tf.contrib.rnn.MultiRNNCell, '__deepcopy__', lambda self, _: self)
class Seq2SeqModel(object):
"""Sequence-to-sequence model with attention and for multiple buckets.
This class implements a multi-layer recurrent neural network as encoder,
this paper: http://arxiv.org/abs/1412.7449 - please look there for details,
or into the seq2seq library for complete model implementation.
This class also allows to use GRU cells in addition to LSTM cells, and
version of this model, but with bi-directional encoder, was presented in
http://arxiv.org/abs/1409.0473
and sampled softmax is described in Section 3 of the following paper.
http://arxiv.org/abs/1412.2007
"""
def __init__(self,
source_vocab_size,
target_vocab_size,
buckets,
size,
num_layers,
max_gradient_norm,
batch_size,
learning_rate,
learning_rate_decay_factor,
use_lstm=False,
num_samples=512,
forward_only=False,
dtype=tf.float32):
"""Create the model.
Args:
source_vocab_size: size of the source vocabulary.
target_vocab_size: size of the target vocabulary.
buckets: a list of pairs (I, O), where I specifies maximum input length
that will be processed in that bucket, and O specifies maximum output
longer than O will be pushed to the next bucket and padded accordingly.
We assume that the list is sorted, e.g., [(2, 4), (8, 16)].
size: number of units in each layer of the model.
num_layers: number of layers in the model.
max_gradient_norm: gradients will be clipped to maximally this norm.
batch_size: the size of the batches used during training;
the model construction is independent of batch_size, so it can be
changed after initialization if this is convenient, e.g., for decoding.
learning_rate: learning rate to start with.
learning_rate_decay_factor: decay learning rate by this much when needed.
use_lstm: if true, we use LSTM cells instead of GRU cells.
forward_only: if set, we do not construct the backward pass in the model.
dtype: the data type to use to store internal variables.
"""
self.source_vocab_size = source_vocab_size
self.target_vocab_size = target_vocab_size
self.buckets = buckets
self.batch_size = batch_size
self.learning_rate = tf.Variable(
float(learning_rate), trainable=False, dtype=dtype)
self.learning_rate_decay_op = self.learning_rate.assign(
self.learning_rate * learning_rate_decay_factor)
self.global_step = tf.Variable(0, trainable=False)
# If we use sampled softmax, we need an output projection.
output_projection = None
softmax_loss_function = None
# Sampled softmax only makes sense if we sample less than vocabulary size.
if num_samples > 0 and num_samples < self.target_vocab_size:
w_t = tf.get_variable("proj_w", [self.target_vocab_size, size], dtype=dtype)
w = tf.transpose(w_t)
b = tf.get_variable("proj_b", [self.target_vocab_size], dtype=dtype)
output_projection = (w, b)
def sampled_loss(labels, logits):
labels = tf.reshape(labels, [-1, 1])
# We need to compute the sampled_softmax_loss using 32bit floats to
# avoid numerical instabilities.
local_w_t = tf.cast(w_t, tf.float32)
local_b = tf.cast(b, tf.float32)
local_inputs = tf.cast(logits, tf.float32)
return tf.cast(
tf.nn.sampled_softmax_loss(
weights=local_w_t,
biases=local_b,
labels=labels,
inputs=local_inputs,
num_sampled=num_samples,
num_classes=self.target_vocab_size),
dtype)
softmax_loss_function = sampled_loss
# Create the internal multi-layer cell for our RNN.
def single_cell():
return tf.contrib.rnn.GRUCell(size)
if use_lstm:
def single_cell():
return tf.contrib.rnn.BasicLSTMCell(size)
cell = single_cell()
if num_layers > 1:
cell = tf.contrib.rnn.MultiRNNCell([single_cell() for _ in range(num_layers)])
# The seq2seq function: we use embedding for the input and attention.
def seq2seq_f(encoder_inputs, decoder_inputs, do_decode):
return tf.contrib.legacy_seq2seq.embedding_attention_seq2seq(
encoder_inputs,
decoder_inputs,
cell,
num_encoder_symbols=source_vocab_size,
num_decoder_symbols=target_vocab_size,
embedding_size=size,
output_projection=output_projection,
feed_previous=do_decode,
dtype=dtype)
self.encoder_inputs = []
self.decoder_inputs = []
self.target_weights = []
for i in xrange(buckets[-1][0]): # Last bucket is the biggest one.
self.encoder_inputs.append(tf.placeholder(tf.int32, shape=[None],
name="encoder{0}".format(i)))
for i in xrange(buckets[-1][1] + 1): # 因为增加了“go”标志
self.decoder_inputs.append(tf.placeholder(tf.int32, shape=[None],
name="decoder{0}".format(i)))
self.target_weights.append(tf.placeholder(dtype, shape=[None],
name="weight{0}".format(i)))
targets = [self.decoder_inputs[i + 1]
for i in xrange(len(self.decoder_inputs) - 1)]
# Training outputs and losses.
'''
lambda x, y的x是指encoder_inputs,y是指decoder_inputs
'''
if forward_only:
self.outputs, self.losses = tf.contrib.legacy_seq2seq.model_with_buckets(
self.encoder_inputs, self.decoder_inputs, targets,
self.target_weights, buckets, lambda x, y: seq2seq_f(x, y, True),
softmax_loss_function=softmax_loss_function)
# If we use output projection, we need to project outputs for decoding.
if output_projection is not None:
for b in xrange(len(buckets)):
self.outputs[b] = [
tf.matmul(output, output_projection[0]) + output_projection[1]
for output in self.outputs[b]
]
else:
self.outputs, self.losses = tf.contrib.legacy_seq2seq.model_with_buckets(
self.encoder_inputs, self.decoder_inputs, targets,
self.target_weights, buckets,
lambda x, y: seq2seq_f(x, y, False),
softmax_loss_function=softmax_loss_function)
# Gradients and SGD update operation for training the model.
params = tf.trainable_variables()
if not forward_only:
self.gradient_norms = []
self.updates = []
opt = tf.train.GradientDescentOptimizer(self.learning_rate)
for b in xrange(len(buckets)):
gradients = tf.gradients(self.losses[b], params)
clipped_gradients, norm = tf.clip_by_global_norm(gradients,
max_gradient_norm)
self.gradient_norms.append(norm)
self.updates.append(opt.apply_gradients(
zip(clipped_gradients, params), global_step=self.global_step))
self.saver = tf.train.Saver(tf.global_variables())
def step(self, session, encoder_inputs, decoder_inputs, target_weights,
bucket_id, forward_only):
"""Run a step of the model feeding the given inputs.
Args:
session: tensorflow session to use.
encoder_inputs: list of numpy int vectors to feed as encoder inputs.
decoder_inputs: list of numpy int vectors to feed as decoder inputs.
target_weights: list of numpy float vectors to feed as target weights.
bucket_id: which bucket of the model to use.
forward_only: whether to do the backward step or only forward.
Returns:
A triple consisting of gradient norm (or None if we did not do backward),
average perplexity, and the outputs.
Raises:
ValueError: if length of encoder_inputs, decoder_inputs, or
target_weights disagrees with bucket size for the specified bucket_id.
"""
# Check if the sizes match.
encoder_size, decoder_size = self.buckets[bucket_id]
#encoder_inputs的shape为(encoder_size,batch_size)
if len(encoder_inputs) != encoder_size:
raise ValueError("Encoder length must be equal to the one in bucket,"
" %d != %d." % (len(encoder_inputs), encoder_size))
if len(decoder_inputs) != decoder_size:
raise ValueError("Decoder length must be equal to the one in bucket,"
" %d != %d." % (len(decoder_inputs), decoder_size))
if len(target_weights) != decoder_size:
raise ValueError("Weights length must be equal to the one in bucket,"
" %d != %d." % (len(target_weights), decoder_size))
# Input feed: encoder inputs, decoder inputs, target_weights, as provided.
input_feed = {}
for k in xrange(encoder_size):
input_feed[self.encoder_inputs[k].name] = encoder_inputs[k]
for k in xrange(decoder_size):
input_feed[self.decoder_inputs[k].name] = decoder_inputs[k]
input_feed[self.target_weights[k].name] = target_weights[k]
# Since our targets are decoder inputs shifted by one, we need one more.
last_target = self.decoder_inputs[decoder_size].name
input_feed[last_target] = np.zeros([self.batch_size], dtype=np.int32)
# Output feed: depends on whether we do a backward step or not.
if not forward_only:
output_feed = [self.updates[bucket_id], # Update Op that does SGD.
self.gradient_norms[bucket_id], # Gradient norm.
self.losses[bucket_id]] # Loss for this batch.
else:
output_feed = [self.losses[bucket_id]] # Loss for this batch.
for l in xrange(decoder_size): # Output logits.
output_feed.append(self.outputs[bucket_id][l])
outputs = session.run(output_feed, input_feed)
if not forward_only:
return outputs[1], outputs[2], None # Gradient norm, loss, no outputs.
else:
return None, outputs[0], outputs[1:] # No gradient norm, loss, outputs.
'''
根据指定bucket_id,产生batch_encoder_inputs和batch_decoder_inputs
这里batch_encoder_inputs和batch_decoder_inputs的shape都由原来的(batch_size,encoder_size)
变为(encoder_size,batch_size),方便进行batch训练
'''
def get_batch(self, data, bucket_id):
"""Get a random batch of data from the specified bucket, prepare for step.
To feed data in step(..) it must be a list of batch-major vectors, while
function is to re-index data cases to be in the proper format for feeding.
Args:
data: a tuple of size len(self.buckets) in which each element contains
lists of pairs of input and output data that we use to create a batch.
bucket_id: integer, which bucket to get the batch for.
Returns:
The triple (encoder_inputs, decoder_inputs, target_weights) for
the constructed batch that has the proper format to call step(...) later.
"""
encoder_size, decoder_size = self.buckets[bucket_id]
encoder_inputs, decoder_inputs = [], []
# Get a random batch of encoder and decoder inputs from data,
# pad them if needed, reverse encoder inputs and add GO to decoder.
for _ in xrange(self.batch_size):
encoder_input, decoder_input = random.choice(data[bucket_id])
# Encoder inputs are padded and then reversed.
encoder_pad = [data_utils.PAD_ID] * (encoder_size - len(encoder_input))
encoder_inputs.append(list(reversed(encoder_input + encoder_pad)))
# Decoder inputs get an extra "GO" symbol, and are padded then.
decoder_pad_size = decoder_size - len(decoder_input) - 1
decoder_inputs.append([data_utils.GO_ID] + decoder_input +
[data_utils.PAD_ID] * decoder_pad_size)
# Now we create batch-major vectors from the data selected above.
batch_encoder_inputs, batch_decoder_inputs, batch_weights = [], [], []
# Batch encoder inputs are just re-indexed encoder_inputs.
#encoder_inputs的shape为(batch_size,encoder_size)
#batch_encoder_inputs的shape为(encoder_size,batch_size)
for length_idx in xrange(encoder_size):
batch_encoder_inputs.append(
np.array([encoder_inputs[batch_idx][length_idx]
for batch_idx in xrange(self.batch_size)], dtype=np.int32))
# Batch decoder inputs are re-indexed decoder_inputs, we create weights.
for length_idx in xrange(decoder_size):
batch_decoder_inputs.append(
np.array([decoder_inputs[batch_idx][length_idx]
for batch_idx in xrange(self.batch_size)], dtype=np.int32))
# Create target_weights to be 0 for targets that are padding.
batch_weight = np.ones(self.batch_size, dtype=np.float32)
for batch_idx in xrange(self.batch_size):
# We set weight to 0 if the corresponding target is a PAD symbol.
# The corresponding target is decoder_input shifted by 1 forward.
if length_idx < decoder_size - 1:
target = decoder_inputs[batch_idx][length_idx + 1]
#如果到底decoder的最后一个单词或target为pad,则不需要比较,即不考虑这一部分的损失函数
if length_idx == decoder_size - 1 or target == data_utils.PAD_ID:
batch_weight[batch_idx] = 0.0
batch_weights.append(batch_weight) #shape为(encoder_size,batch_size)
return batch_encoder_inputs, batch_decoder_inputs, batch_weights