tf.train.shuffle_batch函数解析

tf.train.shuffle_batch函数解析

觉得有用的话,欢迎一起讨论相互学习~

tf.train.shuffle_batch

• (tensor_list, batch_size, capacity, min_after_dequeue, num_threads=1, seed=None, enqueue_many=False, shapes=None, name=None)

• Creates batches by randomly shuffling tensors. 通过随机打乱张量的顺序创建批次.

简单来说就是读取一个文件并且加载一个张量中的batch_size行

This function adds the following to the current Graph:
这个函数将以下内容加入到现有的图中.

• A shuffling queue into which tensors from tensor_list are enqueued.
  一个由传入张量组成的随机乱序队列

• A dequeue_many operation to create batches from the queue.
  从张量队列中取出张量的出队操作

• A QueueRunner to QUEUE_RUNNER collection, to enqueue the tensors
  from tensor_list.
  一个队列运行器管理出队操作.
  If enqueue_many is False, tensor_list is assumed to represent a single example. An input tensor with shape [x, y, z] will be output as a tensor with shape [batch_size, x, y, z].

• If enqueue_many is True, tensor_list is assumed to represent a batch of examples, where the first dimension is indexed by example, and all members of tensor_list should have the same size in the first dimension. If an input tensor has shape [*, x, y, z], the output will have shape [batch_size, x, y, z].

enqueue_many主要是设置tensor中的数据是否能重复,如果想要实现同一个样本多次出现可以将其设置为:“True”,如果只想要其出现一次,也就是保持数据的唯一性,这时候我们将其设置为默认值:“False”

• The capacity argument controls the how long the prefetching is allowed to grow the queues. capacity控制了预抓取操作对于增加队列长度操作的长度.

• For example:

# Creates batches of 32 images and 32 labels.
image_batch, label_batch = tf.train.shuffle_batch( [single_image, single_label], batch_size=32, num_threads=4,capacity=50000,min_after_dequeue=10000)

这段代码写的是从[single_image, single_label]利用4个线程读取32个数据作为一个batch

Args:

• tensor_list: The list of tensors to enqueue.
  入队的张量列表
• batch_size: The new batch size pulled from the queue.
  表示进行一次批处理的tensors数量.
• capacity: An integer. The maximum number of elements in the queue.

容量:一个整数,队列中的最大的元素数.
这个参数一定要比min_after_dequeue参数的值大,并且决定了我们可以进行预处理操作元素的最大值.
推荐其值为:
$$capacity=(min\_after\_dequeue+(num\_threads+asmallsafetymargin\ast batc{h}_{s}ize)$$

• min_after_dequeue: Minimum number elements in the queue after a
  dequeue(出列), used to ensure a level of mixing of elements.
• 当一次出列操作完成后,队列中元素的最小数量,往往用于定义元素的混合级别.
• 定义了随机取样的缓冲区大小,此参数越大表示更大级别的混合但是会导致启动更加缓慢,并且会占用更多的内存
• num_threads: The number of threads enqueuing tensor_list.
• 设置num_threads的值大于1,使用多个线程在tensor_list中读取文件,这样保证了同一时刻只在一个文件中进行读取操作(但是读取速度依然优于单线程),而不是之前的同时读取多个文件,这种方案的优点是:

1. 避免了两个不同的线程从同一文件中读取用一个样本
2. 避免了过多的磁盘操作

• seed: Seed for the random shuffling within the queue.
  打乱tensor队列的随机数种子
• enqueue_many: Whether each tensor in tensor_list is a single example.
  定义tensor_list中的tensor是否冗余.
• shapes: (Optional) The shapes for each example. Defaults to the
  inferred shapes for tensor_list.
  用于改变读取tensor的形状,默认情况下和直接读取的tensor的形状一致.
• name: (Optional) A name for the operations.

Returns:

• A list of tensors with the same number and types as tensor_list.
  默认返回一个和读取tensor_list数据和类型一个tensor列表.