gradient clipping

源链接：http://blog.csdn.net/u014595019/article/details/52759104

损失函数计算

# softmax_w , shape=[hidden_size, vocab_size], 用于将distributed表示的单词转化为one-hot表示
softmax_w = tf.get_variable(
    "softmax_w", [size, vocab_size], dtype=data_type())
softmax_b = tf.get_variable("softmax_b", [vocab_size], dtype=data_type())
# [batch*numsteps, vocab_size] 从隐藏语义转化成完全表示
logits = tf.matmul(output, softmax_w) + softmax_b

# loss , shape=[batch*num_steps]
# 带权重的交叉熵计算
loss = tf.nn.seq2seq.sequence_loss_by_example(
    [logits],   # output [batch*numsteps, vocab_size]
    [tf.reshape(self._targets, [-1])],  # target, [batch_size, num_steps] 然后展开成一维【列表】
    [tf.ones([batch_size * num_steps], dtype=data_type())]) # weight
self._cost = cost = tf.reduce_sum(loss) / batch_size # 计算得到平均每批batch的误差
self._final_state = state
self._lr = tf.Variable(0.0, trainable=False)  # lr 指的是 learning_rate
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(cost, tvars),
                                    config.max_grad_norm)
# 梯度下降优化，指定学习速率
optimizer = tf.train.GradientDescentOptimizer(self._lr)
# optimizer = tf.train.AdamOptimizer()
# optimizer = tf.train.GradientDescentOptimizer(0.5)
self._train_op = optimizer.apply_gradients(zip(grads, tvars))  # 将梯度应用于变量
# self._train_op = optimizer.minimize(grads)

全部源代码:

# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# 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.
# ==============================================================================

"""Example / benchmark for building a PTB LSTM model.
Trains the model described in:
(Zaremba, et. al.) Recurrent Neural Network Regularization
http://arxiv.org/abs/1409.2329
There are 3 supported model configurations:
===========================================
| config | epochs | train | valid  | test
===========================================
| small  | 13     | 37.99 | 121.39 | 115.91
| medium | 39     | 48.45 |  86.16 |  82.07
| large  | 55     | 37.87 |  82.62 |  78.29
The exact results may vary depending on the random initialization.
The hyperparameters used in the model:
- init_scale - the initial scale of the weights
- learning_rate - the initial value of the learning rate
- max_grad_norm - the maximum permissible norm of the gradient
- num_layers - the number of LSTM layers
- num_steps - the number of unrolled steps of LSTM
- hidden_size - the number of LSTM units
- max_epoch - the number of epochs trained with the initial learning rate
- max_max_epoch - the total number of epochs for training
- keep_prob - the probability of keeping weights in the dropout layer
- lr_decay - the decay of the learning rate for each epoch after "max_epoch"
- batch_size - the batch size
The data required for this example is in the data/ dir of the
PTB dataset from Tomas Mikolov's webpage:
$ wget http://www.fit.vutbr.cz/~imikolov/rnnlm/simple-examples.tgz
$ tar xvf simple-examples.tgz
To run:
$ python ptb_word_lm.py --data_path=simple-examples/data/
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import time

import numpy as np
import tensorflow as tf

from tensorflow.models.rnn.ptb import reader

flags = tf.flags
logging = tf.logging

flags.DEFINE_string(
    "model", "small",
    "A type of model. Possible options are: small, medium, large.")
flags.DEFINE_string("data_path", '/home/multiangle/download/simple-examples/data/', "data_path")
flags.DEFINE_bool("use_fp16", False,
                  "Train using 16-bit floats instead of 32bit floats")

FLAGS = flags.FLAGS


def data_type():
    return tf.float16 if FLAGS.use_fp16 else tf.float32


class PTBModel(object):
    """The PTB model."""

    def __init__(self, is_training, config):
        """
        :param is_training: 是否要进行训练.如果is_training=False,则不会进行参数的修正。
        """
        self.batch_size = batch_size = config.batch_size
        self.num_steps = num_steps = config.num_steps
        size = config.hidden_size
        vocab_size = config.vocab_size

        self._input_data = tf.placeholder(tf.int32, [batch_size, num_steps])    # 输入
        self._targets = tf.placeholder(tf.int32, [batch_size, num_steps])       # 预期输出，两者都是index序列，长度为num_step

        # Slightly better results can be obtained with forget gate biases
        # initialized to 1 but the hyperparameters of the model would need to be
        # different than reported in the paper.
        lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(size, forget_bias=0.0, state_is_tuple=True)
        if is_training and config.keep_prob < 1: # 在外面包裹一层dropout
            lstm_cell = tf.nn.rnn_cell.DropoutWrapper(
                lstm_cell, output_keep_prob=config.keep_prob)
        cell = tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * config.num_layers, state_is_tuple=True) # 多层lstm cell 堆叠起来

        self._initial_state = cell.zero_state(batch_size, data_type()) # 参数初始化,rnn_cell.RNNCell.zero_state

        with tf.device("/cpu:0"):
            embedding = tf.get_variable(
                "embedding", [vocab_size, size], dtype=data_type()) # vocab size * hidden size, 将单词转成embedding描述
            # 将输入seq用embedding表示, shape=[batch, steps, hidden_size]
            inputs = tf.nn.embedding_lookup(embedding, self._input_data)

        if is_training and config.keep_prob < 1:
            inputs = tf.nn.dropout(inputs, config.keep_prob)

        # Simplified version of tensorflow.models.rnn.rnn.py's rnn().
        # This builds an unrolled LSTM for tutorial purposes only.
        # In general, use the rnn() or state_saving_rnn() from rnn.py.
        #
        # The alternative version of the code below is:
        #
        # inputs = [tf.squeeze(input_, [1])
        #           for input_ in tf.split(1, num_steps, inputs)]
        # outputs, state = tf.nn.rnn(cell, inputs, initial_state=self._initial_state)
        outputs = []
        state = self._initial_state # state 表示 各个batch中的状态
        with tf.variable_scope("RNN"):
            for time_step in range(num_steps):
                if time_step > 0: tf.get_variable_scope().reuse_variables()
                # cell_out: [batch, hidden_size]
                (cell_output, state) = cell(inputs[:, time_step, :], state)
                outputs.append(cell_output)  # output: shape[num_steps][batch,hidden_size]

        # 把之前的list展开，成[batch, hidden_size*num_steps],然后 reshape, 成[batch*numsteps, hidden_size]
        output = tf.reshape(tf.concat(1, outputs), [-1, size])

        # softmax_w , shape=[hidden_size, vocab_size], 用于将distributed表示的单词转化为one-hot表示
        softmax_w = tf.get_variable(
            "softmax_w", [size, vocab_size], dtype=data_type())
        softmax_b = tf.get_variable("softmax_b", [vocab_size], dtype=data_type())
        # [batch*numsteps, vocab_size] 从隐藏语义转化成完全表示
        logits = tf.matmul(output, softmax_w) + softmax_b

        # loss , shape=[batch*num_steps]
        # 带权重的交叉熵计算
        loss = tf.nn.seq2seq.sequence_loss_by_example(
            [logits],   # output [batch*numsteps, vocab_size]
            [tf.reshape(self._targets, [-1])],  # target, [batch_size, num_steps] 然后展开成一维【列表】
            [tf.ones([batch_size * num_steps], dtype=data_type())]) # weight
        self._cost = cost = tf.reduce_sum(loss) / batch_size # 计算得到平均每批batch的误差
        self._final_state = state

        if not is_training:  # 如果没有训练，则不需要更新state的值。
            return

        self._lr = tf.Variable(0.0, trainable=False)
        tvars = tf.trainable_variables()
        # clip_by_global_norm: 梯度衰减，具体算法为t_list[i] * clip_norm / max(global_norm, clip_norm)
        # 这里gradients求导，ys和xs都是张量
        # 返回一个长为len(xs)的张量，其中的每个元素都是\grad{\frac{dy}{dx}}
        # clip_by_global_norm 用于控制梯度膨胀,前两个参数t_list, global_norm, 则
        # t_list[i] * clip_norm / max(global_norm, clip_norm)
        # 其中 global_norm = sqrt(sum([l2norm(t)**2 for t in t_list]))
        grads, _ = tf.clip_by_global_norm(tf.gradients(cost, tvars),
                                          config.max_grad_norm)

        # 梯度下降优化，指定学习速率
        optimizer = tf.train.GradientDescentOptimizer(self._lr)
        # optimizer = tf.train.AdamOptimizer()
        # optimizer = tf.train.GradientDescentOptimizer(0.5)
        self._train_op = optimizer.apply_gradients(zip(grads, tvars))  # 将梯度应用于变量

        self._new_lr = tf.placeholder(
            tf.float32, shape=[], name="new_learning_rate")     #   用于外部向graph输入新的 lr值
        self._lr_update = tf.assign(self._lr, self._new_lr)     #   使用new_lr来更新lr的值

    def assign_lr(self, session, lr_value):
        # 使用 session 来调用 lr_update 操作
        session.run(self._lr_update, feed_dict={self._new_lr: lr_value})

    @property
    def input_data(self):
        return self._input_data

    @property
    def targets(self):
        return self._targets

    @property
    def initial_state(self):
        return self._initial_state

    @property
    def cost(self):
        return self._cost

    @property
    def final_state(self):
        return self._final_state

    @property
    def lr(self):
        return self._lr

    @property
    def train_op(self):
        return self._train_op


class SmallConfig(object):
    """Small config."""
    init_scale = 0.1        #
    learning_rate = 1.0     # 学习速率
    max_grad_norm = 5       # 用于控制梯度膨胀，
    num_layers = 2          # lstm层数
    num_steps = 20          # 单个数据中，序列的长度。
    hidden_size = 200       # 隐藏层规模
    max_epoch = 4           # epoch<max_epoch时，lr_decay值=1,epoch>max_epoch时,lr_decay逐渐减小
    max_max_epoch = 13      # 指的是整个文本循环13遍。
    keep_prob = 1.0
    lr_decay = 0.5          # 学习速率衰减
    batch_size = 20         # 每批数据的规模，每批有20个。
    vocab_size = 10000      # 词典规模，总共10K个词


class MediumConfig(object):
    """Medium config."""
    init_scale = 0.05
    learning_rate = 1.0
    max_grad_norm = 5
    num_layers = 2
    num_steps = 35
    hidden_size = 650
    max_epoch = 6
    max_max_epoch = 39
    keep_prob = 0.5
    lr_decay = 0.8
    batch_size = 20
    vocab_size = 10000


class LargeConfig(object):
    """Large config."""
    init_scale = 0.04
    learning_rate = 1.0
    max_grad_norm = 10
    num_layers = 2
    num_steps = 35
    hidden_size = 1500
    max_epoch = 14
    max_max_epoch = 55
    keep_prob = 0.35
    lr_decay = 1 / 1.15
    batch_size = 20
    vocab_size = 10000


class TestConfig(object):
    """Tiny config, for testing."""
    init_scale = 0.1
    learning_rate = 1.0
    max_grad_norm = 1
    num_layers = 1
    num_steps = 2
    hidden_size = 2
    max_epoch = 1
    max_max_epoch = 1
    keep_prob = 1.0
    lr_decay = 0.5
    batch_size = 20
    vocab_size = 10000


def run_epoch(session, model, data, eval_op, verbose=False):
    """Runs the model on the given data."""
    # epoch_size 表示批次总数。也就是说，需要向session喂这么多次数据
    epoch_size = ((len(data) // model.batch_size) - 1) // model.num_steps  # // 表示整数除法
    start_time = time.time()
    costs = 0.0
    iters = 0
    state = session.run(model.initial_state)
    for step, (x, y) in enumerate(reader.ptb_iterator(data, model.batch_size,
                                                      model.num_steps)):
        fetches = [model.cost, model.final_state, eval_op] # 要进行的操作，注意训练时和其他时候eval_op的区别
        feed_dict = {}      # 设定input和target的值
        feed_dict[model.input_data] = x
        feed_dict[model.targets] = y
        for i, (c, h) in enumerate(model.initial_state):
            feed_dict[c] = state[i].c   # 这部分有什么用？看不懂
            feed_dict[h] = state[i].h
        cost, state, _ = session.run(fetches, feed_dict) # 运行session,获得cost和state
        costs += cost   # 将 cost 累积
        iters += model.num_steps

        if verbose and step % (epoch_size // 10) == 10:  # 也就是每个epoch要输出10个perplexity值
            print("%.3f perplexity: %.3f speed: %.0f wps" %
                  (step * 1.0 / epoch_size, np.exp(costs / iters),
                   iters * model.batch_size / (time.time() - start_time)))

    return np.exp(costs / iters)


def get_config():
    if FLAGS.model == "small":
        return SmallConfig()
    elif FLAGS.model == "medium":
        return MediumConfig()
    elif FLAGS.model == "large":
        return LargeConfig()
    elif FLAGS.model == "test":
        return TestConfig()
    else:
        raise ValueError("Invalid model: %s", FLAGS.model)


# def main(_):
if __name__=='__main__':
    if not FLAGS.data_path:
        raise ValueError("Must set --data_path to PTB data directory")
    print(FLAGS.data_path)

    raw_data = reader.ptb_raw_data(FLAGS.data_path) # 获取原始数据
    train_data, valid_data, test_data, _ = raw_data

    config = get_config()
    eval_config = get_config()
    eval_config.batch_size = 1
    eval_config.num_steps = 1

    with tf.Graph().as_default(), tf.Session() as session:
        initializer = tf.random_uniform_initializer(-config.init_scale, # 定义如何对参数变量初始化
                                                    config.init_scale)
        with tf.variable_scope("model", reuse=None,initializer=initializer):
            m = PTBModel(is_training=True, config=config)   # 训练模型， is_trainable=True
        with tf.variable_scope("model", reuse=True,initializer=initializer):
            mvalid = PTBModel(is_training=False, config=config) #  交叉检验和测试模型，is_trainable=False
            mtest = PTBModel(is_training=False, config=eval_config)

        summary_writer = tf.train.SummaryWriter('/tmp/lstm_logs',session.graph)

        tf.initialize_all_variables().run()  # 对参数变量初始化

        for i in range(config.max_max_epoch):   # 所有文本要重复多次进入模型训练
            # learning rate 衰减
            # 在 遍数小于max epoch时， lr_decay = 1 ; > max_epoch时， lr_decay = 0.5^(i-max_epoch)
            lr_decay = config.lr_decay ** max(i - config.max_epoch, 0.0)
            m.assign_lr(session, config.learning_rate * lr_decay) # 设置learning rate

            print("Epoch: %d Learning rate: %.3f" % (i + 1, session.run(m.lr)))
            train_perplexity = run_epoch(session, m, train_data, m.train_op,verbose=True) # 训练困惑度
            print("Epoch: %d Train Perplexity: %.3f" % (i + 1, train_perplexity))
            valid_perplexity = run_epoch(session, mvalid, valid_data, tf.no_op()) # 检验困惑度
            print("Epoch: %d Valid Perplexity: %.3f" % (i + 1, valid_perplexity))

        test_perplexity = run_epoch(session, mtest, test_data, tf.no_op())  # 测试困惑度
        print("Test Perplexity: %.3f" % test_perplexity)


# if __name__ == "__main__":
#     tf.app.run()