TensorFlow笔记(基础篇):加载数据之从队列中读取

前言

整体步骤

在TensorFlow中进行模型训练时,在官网给出的三种读取方式,中最好的文件读取方式就是将利用队列进行文件读取,而且步骤有两步:
1. 把样本数据写入TFRecords二进制文件
2. 从队列中读取数据

读取TFRecords文件步骤

使用队列读取数TFRecords 文件 数据的步骤
1. 创建张量,从二进制文件读取一个样本数据
2. 创建张量,从二进制文件随机读取一个mini-batch
3. 把每一批张量传入网络作为输入点

TensorFlow使用TFRecords文件训练样本的步骤

1. 在生成文件名的序列中,设定epoch数量
2. 训练时,设定为无穷循环
3. 在读取数据时,如果捕捉到错误,终止

source code

tensorflow-master ensorflowexampleshow_tos eading_datafully_connected_reader.py(1.2.1)

CODE

代码与解析

解析主要在注释中,最后一个模块if __name__ == '__main__':的运行,建议参考’http://blog.csdn.net/fontthrone/article/details/76735591’

import tensorflow as tf
import os

# from tensorflow.contrib.learn.python.learn.datasets import mnist
# 注意上面的这个mnist 与 example 中的 mnist 是不同的,本文件中请使用下面的那个 mnist

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import argparse
import os.path
import sys
import time

from tensorflow.examples.tutorials.mnist import mnist

# Basic model parameters as external flags.
FLAGS = None

# This part of the code is added by FontTian,which comes from the source code of tensorflow.examples.tutorials.mnist
# The MNIST images are always 28x28 pixels.
# IMAGE_SIZE = 28
# IMAGE_PIXELS = IMAGE_SIZE * IMAGE_SIZE

# Constants used for dealing with the files, matches convert_to_records.
TRAIN_FILE = 'train.tfrecords'
VALIDATION_FILE = 'validation.tfrecords'


def read_and_decode(filename_queue):
    reader = tf.TFRecordReader()
    _, serialized_example = reader.read(filename_queue)
    features = tf.parse_single_example(
        serialized_example,
        # Defaults are not specified since both keys are required.
        # 必须写明faetures 中的 key 的名称
        features={
            'image_raw': tf.FixedLenFeature([], tf.string),
            'label': tf.FixedLenFeature([], tf.int64),
        })

    # Convert from a scalar string tensor (whose single string has
    # length mnist.IMAGE_PIXELS) to a uint8 tensor with shape
    # [mnist.IMAGE_PIXELS].
    # 将一个标量字符串张量(其单个字符串的长度是mnist.image像素) # 0 维的Tensor
    # 转换为一个带有形状mnist.图像像素的uint8张量。 # 一维的Tensor
    image = tf.decode_raw(features['image_raw'], tf.uint8)
    # print(tf.shape(image)) # Tensor("input/Shape:0", shape=(1,), dtype=int32)

    image.set_shape([mnist.IMAGE_PIXELS])
    # print(tf.shape(image)) # Tensor("input/Shape_1:0", shape=(1,), dtype=int32)

    # OPTIONAL: Could reshape into a 28x28 image and apply distortions
    # here.  Since we are not applying any distortions in this
    # example, and the next step expects the image to be flattened
    # into a vector, we don't bother.

    # Convert from [0, 255] -> [-0.5, 0.5] floats.
    image = tf.cast(image, tf.float32) * (1. / 255) - 0.5
    # print(tf.shape(image)) # Tensor("input/Shape_2:0", shape=(1,), dtype=int32)

    # Convert label from a scalar uint8 tensor to an int32 scalar.
    label = tf.cast(features['label'], tf.int32)
    # print(tf.shape(label)) # Tensor("input/Shape_3:0", shape=(0,), dtype=int32)

    return image, label


# 使用 tf.train.shuffle_batch 将前面生成的样本随机化,获得一个最小批次的张量
def inputs(train, batch_size, num_epochs):
    """Reads input data num_epochs times.

    Args:
      train: Selects between the training (True) and validation (False) data.
      batch_size: Number of examples per returned batch.
      num_epochs: Number of times to read the input data, or 0/None to
         train forever.

    Returns:
      A tuple (images, labels), where:
      * images is a float tensor with shape [batch_size, mnist.IMAGE_PIXELS]
        in the range [-0.5, 0.5].
      * labels is an int32 tensor with shape [batch_size] with the true label,
        a number in the range [0, mnist.NUM_CLASSES).
      Note that an tf.train.QueueRunner is added to the graph, which
      must be run using e.g. tf.train.start_queue_runners().

    输入参数:
      train: Selects between the training (True) and validation (False) data.
      batch_size: 训练的每一批有多少个样本
      num_epochs: 读取输入数据的次数, or 0/None 表示永远训练下去

    返回结果:
      A tuple (images, labels), where:
      * images is a float tensor with shape [batch_size, mnist.IMAGE_PIXELS]
        范围: [-0.5, 0.5].
      * labels is an int32 tensor with shape [batch_size] with the true label,
        范围: [0, mnist.NUM_CLASSES).
      注意 :  tf.train.QueueRunner 被添加进 graph, 它必须用 tf.train.start_queue_runners() 来启动线程.

    """

    if not num_epochs: num_epochs = None
    filename = os.path.join(FLAGS.train_dir,
                            TRAIN_FILE if train else VALIDATION_FILE)

    with tf.name_scope('input'):
        # tf.train.string_input_producer 返回一个 QueueRunner,里面有一个 FIFQueue
        filename_queue = tf.train.string_input_producer(
            [filename], num_epochs=num_epochs)
        # 如果样本数据很大,可以分成若干文件,把文件名列表传入

        # Even when reading in multiple threads, share the filename queue.
        image, label = read_and_decode(filename_queue)

        # Shuffle the examples and collect them into batch_size batches.
        # (Internally uses a RandomShuffleQueue.)
        # We run this in two threads to avoid being a bottleneck.
        images, sparse_labels = tf.train.shuffle_batch(
            [image, label], batch_size=batch_size, num_threads=2,
            capacity=1000 + 3 * batch_size,
            # Ensures a minimum amount of shuffling of examples.
            # 留下一部分队列,来保证每次有足够的数据做随机打乱
            min_after_dequeue=1000)

        return images, sparse_labels

def run_training():
    """Train MNIST for a number of steps."""

    # Tell TensorFlow that the model will be built into the default Graph.
    with tf.Graph().as_default():
        # Input images and labels.
        images, labels = inputs(train=True, batch_size=FLAGS.batch_size,
                                num_epochs=FLAGS.num_epochs)

        # 构建一个从推理模型来预测数据的图
        logits = mnist.inference(images,
                                 FLAGS.hidden1,
                                 FLAGS.hidden2)

        # Add to the Graph the loss calculation.
        # 定义损失函数
        loss = mnist.loss(logits, labels)

        # 将模型添加到图操作中
        train_op = mnist.training(loss, FLAGS.learning_rate)

        # 初始化变量的操作
        init_op = tf.group(tf.global_variables_initializer(),
                           tf.local_variables_initializer())

        # Create a session for running operations in the Graph.
        # 在图中创建一个用于运行操作的会话
        sess = tf.Session()

        # 初始化变量,注意:string_input_product 内部创建了一个epoch计数器
        sess.run(init_op)

        # Start input enqueue threads.
        coord = tf.train.Coordinator()
        threads = tf.train.start_queue_runners(sess=sess, coord=coord)

        try:
            step = 0
            while not coord.should_stop():
                start_time = time.time()

                # Run one step of the model.  The return values are
                # the activations from the `train_op` (which is
                # discarded) and the `loss` op.  To inspect the values
                # of your ops or variables, you may include them in
                # the list passed to sess.run() and the value tensors
                # will be returned in the tuple from the call.
                _, loss_value = sess.run([train_op, loss])

                duration = time.time() - start_time

                # Print an overview fairly often.
                if step % 100 == 0:
                    print('Step %d: loss = %.2f (%.3f sec)' % (step, loss_value,
                                                               duration))
                step += 1
        except tf.errors.OutOfRangeError:
            print('Done training for %d epochs, %d steps.' % (FLAGS.num_epochs, step))
        finally:
            # 通知其他线程关闭
            coord.request_stop()

        # Wait for threads to finish.
        coord.join(threads)
        sess.close()

def main(_):
    run_training()


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument(
        '--learning_rate',
        type=float,
        default=0.01,
        help='Initial learning rate.'
    )
    parser.add_argument(
        '--num_epochs',
        type=int,
        default=2,
        help='Number of epochs to run trainer.'
    )
    parser.add_argument(
        '--hidden1',
        type=int,
        default=128,
        help='Number of units in hidden layer 1.'
    )
    parser.add_argument(
        '--hidden2',
        type=int,
        default=32,
        help='Number of units in hidden layer 2.'
    )
    parser.add_argument(
        '--batch_size',
        type=int,
        default=100,
        help='Batch size.'
    )
    parser.add_argument(
        '--train_dir',
        type=str,
        default='/tmp/data',
        help='Directory with the training data.'
    )
    FLAGS, unparsed = parser.parse_known_args()
    tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)

运行结果

Step 0: loss = 2.31 (0.106 sec)
Step 100: loss = 2.14 (0.016 sec)
Step 200: loss = 1.91 (0.016 sec)
Step 300: loss = 1.69 (0.016 sec)
Step 400: loss = 1.28 (0.016 sec)
Step 500: loss = 1.02 (0.016 sec)
Step 600: loss = 0.70 (0.016 sec)
Step 700: loss = 0.71 (0.016 sec)
Step 800: loss = 0.71 (0.016 sec)
Step 900: loss = 0.49 (0.016 sec)
Step 1000: loss = 0.58 (0.016 sec)
Done training for 2 epochs, 1100 steps.

相关

1. 把样本数据写入TFRecords二进制文件 : http://blog.csdn.net/fontthrone/article/details/76727412
2. TensorFlow笔记(基础篇):加载数据之预加载数据与填充数据:http://blog.csdn.net/fontthrone/article/details/76727466
3. python中的argparse模块:http://blog.csdn.net/fontthrone/article/details/76735591