TF之LSTM：基于Tensorflow框架采用PTB数据集建立LSTM网络的自然语言建模

TF之LSTM：基于Tensorflow框架采用PTB数据集建立LSTM网络的自然语言建模

目录

关于PTB数据集

代码实现

关于PTB数据集

PTB （Penn Treebank Dataset）文本数据集是语言模型学习中目前最被广泛使用数据集。
ptb.test.txt    测试集数据文件
ptb.train.txt   训练集数据文件
ptb.valid.txt   验证集数据文件
这三个数据文件中的数据已经经过了预处理，包含了10000 个不同的词语和语句结束标记符（在文本中就是换行符）以及标记稀有词语的特殊符号。
为了让使用PTB数据集更加方便，TensorFlow提供了两个函数来帮助实现数据的预处理。首先，TensorFlow提供了ptb_raw_data函数来读取PTB的原始数据，并将原始数据中的单词转化为单词ID。
训练数据中总共包含了929589 个单词，而这些单词被组成了一个非常长的序列。这个序列通过特殊的标识符给出了每句话结束的位置。在这个数据集中，句子结束的标识符ID为2。
数据集的下载地址：TF的PTB数据集     (别的数据集不匹配的话会出现错误)
 

代码实现

   本代码使用2层 LSTM 网络，且每层有 200 个隐藏单元。在训练中截断的输入序列长度为 32，且使用 Dropout 和梯度截断等方法控制模型的过拟合与梯度爆炸等问题。当简单地训练 3 个 Epoch 后，测试复杂度（Perplexity）降低到了 210，如果多轮训练会更低。

 -*- coding: utf-8 -*-
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
 
import collections
import os
import sys
 
import tensorflow as tf
 
Py3 = sys.version_info[0] == 3
 
def _read_words(filename):
  with tf.gfile.GFile(filename, "r") as f:
    if Py3:
      return f.read().replace("\n", "<eos>").split()
    else:
      return f.read().decode("utf-8").replace("\n", "<eos>").split()
 
 
def _build_vocab(filename):
  data = _read_words(filename)
 
  counter = collections.Counter(data)
  count_pairs = sorted(counter.items(), key=lambda x: (-x[1], x[0]))
 
  words, _ = list(zip(*count_pairs))
  word_to_id = dict(zip(words, range(len(words))))
 
  return word_to_id
 
 
def _file_to_word_ids(filename, word_to_id):
  data = _read_words(filename)
  return [word_to_id[word] for word in data if word in word_to_id]
 
 
def ptb_raw_data(data_path=None):
  """Load PTB raw data from data directory "data_path".
  Reads PTB text files, converts strings to integer ids,
  and performs mini-batching of the inputs.
  The PTB dataset comes from Tomas Mikolov's webpage:
  http://www.fit.vutbr.cz/~imikolov/rnnlm/simple-examples.tgz
  Args:
    data_path: string path to the directory where simple-examples.tgz has
      been extracted.
  Returns:
    tuple (train_data, valid_data, test_data, vocabulary)
    where each of the data objects can be passed to PTBIterator.
  """
 
  train_path = os.path.join(data_path, "ptb.train.txt")
  valid_path = os.path.join(data_path, "ptb.valid.txt")
  test_path = os.path.join(data_path, "ptb.test.txt")
 
  word_to_id = _build_vocab(train_path)
  train_data = _file_to_word_ids(train_path, word_to_id)
  valid_data = _file_to_word_ids(valid_path, word_to_id)
  test_data = _file_to_word_ids(test_path, word_to_id)
  vocabulary = len(word_to_id)
  return train_data, valid_data, test_data, vocabulary
 
 
def ptb_producer(raw_data, batch_size, num_steps, name=None):
  """Iterate on the raw PTB data.
  This chunks up raw_data into batches of examples and returns Tensors that
  are drawn from these batches.
  Args:
    raw_data: one of the raw data outputs from ptb_raw_data.
    batch_size: int, the batch size.
    num_steps: int, the number of unrolls.
    name: the name of this operation (optional).
  Returns:
    A pair of Tensors, each shaped [batch_size, num_steps]. The second element
    of the tuple is the same data time-shifted to the right by one.
  Raises:
    tf.errors.InvalidArgumentError: if batch_size or num_steps are too high.
  """
  with tf.name_scope(name, "PTBProducer", [raw_data, batch_size, num_steps]):
    raw_data = tf.convert_to_tensor(raw_data, name="raw_data", dtype=tf.int32)
 
    data_len = tf.size(raw_data)
    batch_len = data_len // batch_size
    data = tf.reshape(raw_data[0 : batch_size * batch_len],
                      [batch_size, batch_len])
 
    epoch_size = (batch_len - 1) // num_steps
    assertion = tf.assert_positive(
        epoch_size,
        message="epoch_size == 0, decrease batch_size or num_steps")
    with tf.control_dependencies([assertion]):
      epoch_size = tf.identity(epoch_size, name="epoch_size")
 
    i = tf.train.range_input_producer(epoch_size, shuffle=False).dequeue()
    x = tf.strided_slice(data, [0, i * num_steps],
                         [batch_size, (i + 1) * num_steps])
    x.set_shape([batch_size, num_steps])
    y = tf.strided_slice(data, [0, i * num_steps + 1],
                         [batch_size, (i + 1) * num_steps + 1])
    y.set_shape([batch_size, num_steps])
    return x, y

from reader import *
import tensorflow as tf
import numpy as np
 
data_path = 'F:/File_Python/Python_daydayup/data/simple-examples/data'  F:/File_Python/Python_daydayup/data/simple-examples/data
 隐藏层单元数与LSTM层级数
hidden_size = 200
num_layers = 2
词典规模
vocab_size = 10000
 
learning_rate = 1.0
train_batch_size = 16
 训练数据截断长度
train_num_step = 32
 
 在测试时不需要使用截断，测试数据为一个超长序列
eval_batch_size = 1
eval_num_step = 1
num_epoch = 3
结点不被Dropout的概率
keep_prob = 0.5
 
 用于控制梯度爆炸的参数
max_grad_norm = 5
 通过ptbmodel 的类描述模型
class PTBModel(-title class_ inherited__">object):
    def __init__(self, is_training, batch_size, num_steps):
         记录使用的Batch大小和截断长度
        self.batch_size = batch_size
        self.num_steps = num_steps
 
         定义输入层，维度为批量大小×截断长度
        self.input_data = tf.placeholder(tf.int32, [batch_size, num_steps])
         定义预期输出
        self.targets = tf.placeholder(tf.int32, [batch_size, num_steps])
 
         定义使用LSTM结构为循环体，带Dropout的深度RNN
        lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(hidden_size)
        if is_training:
            lstm_cell = tf.nn.rnn_cell.DropoutWrapper(lstm_cell, output_keep_prob=keep_prob)
        cell = tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * num_layers)
 
         初始化状态为0
        self.initial_state = cell.zero_state(batch_size, tf.float32)
 
         将单词ID转换为单词向量，embedding的维度为vocab_size*hidden_size
        embedding = tf.get_variable('embedding', [vocab_size, hidden_size])
         将一个批量内的单词ID转化为词向量，转化后的输入维度为批量大小×截断长度×隐藏单元数
        inputs = tf.nn.embedding_lookup(embedding, self.input_data)
 
         只在训练时使用Dropout
        if is_training: inputs = tf.nn.dropout(inputs, keep_prob)
 
         定义输出列表，这里先将不同时刻LSTM的输出收集起来，再通过全连接层得到最终输出
        outputs = []
         state 储存不同批量中LSTM的状态，初始为0
        state = self.initial_state
        with tf.variable_scope('RNN'):
            for time_step in range(num_steps):
                if time_step > 0: tf.get_variable_scope().reuse_variables()
                 从输入数据获取当前时间步的输入与前一时间步的状态，并传入LSTM结构
                cell_output, state = cell(inputs[:, time_step, :], state)
                 将当前输出加入输出队列
                outputs.append(cell_output)
 
         将输出队列展开成[batch,hidden*num_step]的形状，再reshape为[batch*num_step, hidden]
        output = tf.reshape(tf.concat(outputs, 1), [-1, hidden_size])
 
         将LSTM的输出传入全连接层以生成最后的预测结果。最后结果在每时刻上都是长度为vocab_size的张量
         且经过softmax层后表示下一个位置不同词的概率
        weight = tf.get_variable('weight', [hidden_size, vocab_size])
        bias = tf.get_variable('bias', [vocab_size])
        logits = tf.matmul(output, weight) + bias
 
         定义交叉熵损失函数，一个序列的交叉熵之和
        loss = tf.contrib.legacy_seq2seq.sequence_loss_by_example(
            [logits],   预测的结果
            [tf.reshape(self.targets, [-1])],   期望正确的结果，这里将[batch_size, num_steps]压缩为一维张量
            [tf.ones([batch_size * num_steps], dtype=tf.float32)])   损失的权重，所有为1表明不同批量和时刻的重要程度一样
 
         计算每个批量的平均损失
        self.cost = tf.reduce_sum(loss) / batch_size
        self.final_state = state
 
         只在训练模型时定义反向传播操作
        if not is_training: return
        trainable_variable = tf.trainable_variables()
 
         控制梯度爆炸问题
        grads, _ = tf.clip_by_global_norm(tf.gradients(self.cost, trainable_variable), max_grad_norm)
         如果需要使用Adam作为优化器，可以改为tf.train.AdamOptimizer(learning_rate)，学习率需要降低至0.001左右
        optimizer = tf.train.GradientDescentOptimizer(learning_rate)
         定义训练步骤
        self.train_op = optimizer.apply_gradients(zip(grads, trainable_variable))
def run_epoch(session, model, data, train_op, output_log, epoch_size):
    total_costs = 0.0
    iters = 0
    state = session.run(model.initial_state)
 
      使用当前数据训练或测试模型
    for step in range(epoch_size):
        x, y = session.run(data)
         在当前批量上运行train_op并计算损失值，交叉熵计算的是下一个单词为给定单词的概率
        cost, state, _ = session.run([model.cost, model.final_state, train_op],
                                        {model.input_data: x, model.targets: y, model.initial_state: state})
         将不同时刻和批量的概率就可得到困惑度的对数形式，将这个和做指数运算就可得到困惑度
        total_costs += cost
        iters += model.num_steps
 
         只在训练时输出日志
        if output_log and step % 100 == 0:
            print("After %d steps, perplexity is %.3f" % (step, np.exp(total_costs / iters)))
    return np.exp(total_costs / iters)
def main():
    train_data, valid_data, test_data, _ = ptb_raw_data(data_path)
 
     计算一个epoch需要训练的次数
    train_data_len = len(train_data)
    train_batch_len = train_data_len // train_batch_size
    train_epoch_size = (train_batch_len - 1) // train_num_step
 
    valid_data_len = len(valid_data)
    valid_batch_len = valid_data_len // eval_batch_size
    valid_epoch_size = (valid_batch_len - 1) // eval_num_step
 
    test_data_len = len(test_data)
    test_batch_len = test_data_len // eval_batch_size
    test_epoch_size = (test_batch_len - 1) // eval_num_step
 
    initializer = tf.random_uniform_initializer(-0.05, 0.05)
    with tf.variable_scope("language_model", reuse=None, initializer=initializer):
        train_model = PTBModel(True, train_batch_size, train_num_step)
 
    with tf.variable_scope("language_model", reuse=True, initializer=initializer):
        eval_model = PTBModel(False, eval_batch_size, eval_num_step)
 
     训练模型。
    with tf.Session() as session:
        tf.global_variables_initializer().run()
 
        train_queue = ptb_producer(train_data, train_model.batch_size, train_model.num_steps)