tf.split (API r1.3)

亦凉 2023-02-22 15:55 4阅读 0赞

#  tf.split (API r1.3) #

[https://github.com/tensorflow/docs/blob/r1.3/site/en/api\_docs/api\_docs/python/tf/split.md][https_github.com_tensorflow_docs_blob_r1.3_site_en_api_docs_api_docs_python_tf_split.md]  
`site/en/api_docs/api_docs/python/tf/split.md`

split(
        value,
        num_or_size_splits,
        axis=0,
        num=None,
        name='split'
    )

Defined in `tensorflow/python/ops/array_ops.py`.  
See the guide: Tensor Transformations > Slicing and Joining

Splits a tensor into sub tensors.  
将张量分割成子张量。

If `num_or_size_splits` is an integer type, `num_split`, then splits `value` along dimension `axis` into `num_split` smaller tensors. Requires that `num_split` evenly divides `value.shape[axis]`.  
如果 num\_or\_size\_splits 是整数类型 (num\_split)，则沿维度 axis 分割 value 成为 num\_split 更小的张量。要求 num\_split 均匀分配 value.shape\[axis\]。

If `num_or_size_splits` is not an integer type, it is presumed to be a Tensor `size_splits`, then splits `value` into `len(size_splits)` pieces. The shape of the `i`\-th piece has the same size as the `value` except along dimension `axis` where the size is `size_splits[i]`.  
如果 num\_or\_size\_splits 不是整数类型，则它被认为是一个张量 size\_splits，然后将 value 分割成 len(size\_splits) 块。第 i 部分的形状与 value 的大小相同，除了沿维度 axis 之外的大小 size\_splits\[i\]。

For example:

# 'value' is a tensor with shape [5, 30]
    # Split 'value' into 3 tensors with sizes [4, 15, 11] along dimension 1
    split0, split1, split2 = tf.split(value, [4, 15, 11], 1)
    tf.shape(split0)  # [5, 4]
    tf.shape(split1)  # [5, 15]
    tf.shape(split2)  # [5, 11]
    # Split 'value' into 3 tensors along dimension 1
    split0, split1, split2 = tf.split(value, num_or_size_splits=3, axis=1)
    tf.shape(split0)  # [5, 10]

## 1. Args ##

*  `value`: The `Tensor` to split.
 *  `num_or_size_splits`: Either a 0-D integer `Tensor` indicating the number of splits along split\_dim or a 1-D integer `Tensor` integer tensor containing the sizes of each output tensor along split\_dim. If a scalar then it must evenly divide `value.shape[axis]`; otherwise the sum of sizes along the split dimension must match that of the `value`. (指示沿 split\_dim 分割数量为 0-D 整数 Tensor 或包含沿 split\_dim 每个输出张量大小的 1-D 整数 Tensor。如果为一个标量，那么它必须均匀分割 value.shape\[axis\]，否则沿分割维度的大小总和必须与该 value 相匹配。)
 *  `axis`: A 0-D `int32` `Tensor`. The dimension along which to split. Must be in the range `[-rank(value), rank(value))`. Defaults to 0. (表示分割的尺寸，必须在 \[-rank(value), rank(value)) 范围内，默认为 0。)
 *  `num`: Optional, used to specify the number of outputs when it cannot be inferred from the shape of `size_splits`. (可选的，用于指定无法从 size\_splits 的形状推断出的输出数目。)
 *  `name`: A name for the operation (optional). (操作的名称 (可选)。)

如果 num\_or\_size\_splits 传入的是一个整数，直接在 axis 维度上把张量平均切分成几个小张量。  
如果 num\_or\_size\_splits 传入的是一个向量 (向量各个元素的和要与原本这个维度的数值相等) 就根据这个向量有几个元素分为几项。

## 2. Returns ##

if `num_or_size_splits` is a scalar returns `num_or_size_splits` `Tensor` objects; if `num_or_size_splits` is a 1-D Tensor returns `num_or_size_splits.get_shape[0]` `Tensor` objects resulting from splitting `value`.  
如果 num\_or\_size\_splits 是标量，返回 num\_or\_size\_splits Tensor对象。如果 num\_or\_size\_splits 是一维张量，则返回由 value 分割产生的 num\_or\_size\_splits.get\_shape\[0\] Tensor 对象。

## 3. Raises ##

*  `ValueError`: If `num` is unspecified and cannot be inferred. (如果 num 没有指定并且无法推断。)

## 4. Example ##

## 5. Example ##

## 6. Example ##

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    
    from __future__ import absolute_import
    from __future__ import print_function
    from __future__ import division
    
    import os
    import sys
    import numpy as np
    import tensorflow as tf
    
    sys.path.append(os.path.dirname(os.path.abspath(__file__)))
    current_directory = os.path.dirname(os.path.abspath(__file__))
    
    print(16 * "++--")
    print("current_directory:", current_directory)
    print(16 * "++--")
    
    batch_size = 1
    num_step = 6
    num_input = 2
    
    # x_anchor shape: (batch_size, num_step, num_input)
    x_anchor = tf.constant([[[0, 1],
                             [2, 3],
                             [4, 5],
                             [6, 7],
                             [8, 9],
                             [10, 11]]], dtype=np.float32)
    
    # Permute num_step and batch_size.
    y_anchor = tf.transpose(x_anchor, perm=[1, 0, 2])
    
    # (num_step * batch_size, num_input)
    y_reshape = tf.reshape(y_anchor, [num_step * batch_size, num_input])
    
    # Split data because rnn cell needs a list of inputs for the RNN inner loop num_step * (batch_size, num_input)
    y_split = tf.split(y_reshape, num_step, 0)
    
    with tf.Session() as sess:
        input_x_anchor = sess.run(x_anchor)
        print("type(input_x_anchor):", type(input_x_anchor))
        print("input_x_anchor.shape:", input_x_anchor.shape)
        print(8 * "++--")
    
        output_y_anchor = sess.run(y_anchor)
        print("type(output_y_anchor):", type(output_y_anchor))
        print("output_y_anchor.shape:", output_y_anchor.shape)
        print("output_y_anchor:\n", output_y_anchor)
        print(8 * "++--")
    
        output_y_reshape = sess.run(y_reshape)
        print("type(output_y_reshape):", type(output_y_reshape))
        print("output_y_reshape.shape:", output_y_reshape.shape)
        print("output_y_reshape:\n", output_y_reshape)
        print(8 * "++--")
    
        output_y_split = sess.run(y_split)
        print("type(output_y_split):", type(output_y_split))
        print("output_y_split:\n", output_y_split)
        print(8 * "++--")
    
        print("output_y_split[0-5]:")
        for step in range(num_step):
            print("type(output_y_split[%d]):%s" % (step, type(output_y_split[step])))
            print("output_y_split[%d]:\n" % (step), output_y_split[step])
        print(8 * "++--")
    
        print("output_y_split[0-5]:")
        for step in range(num_step):
            print("type([output_y_split[%d]]):%s" % (step, type([output_y_split[step]])))
            print("[output_y_split[%d]]:\n" % (step), [output_y_split[step]])
        print(8 * "++--")

/usr/bin/python2.7 /home/strong/tensorflow_work/R2CNN_Faster-RCNN_Tensorflow/yongqiang.py
    ++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--
    current_directory: /home/strong/tensorflow_work/R2CNN_Faster-RCNN_Tensorflow
    ++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--
    2019-08-19 10:39:35.983470: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
    2019-08-19 10:39:36.065438: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
    2019-08-19 10:39:36.065680: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
    name: GeForce GTX 1080 major: 6 minor: 1 memoryClockRate(GHz): 1.7335
    pciBusID: 0000:01:00.0
    totalMemory: 7.92GiB freeMemory: 7.39GiB
    2019-08-19 10:39:36.065692: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce GTX 1080, pci bus id: 0000:01:00.0, compute capability: 6.1)
    type(input_x_anchor): <type 'numpy.ndarray'>
    input_x_anchor.shape: (1, 6, 2)
    ++--++--++--++--++--++--++--++--
    type(output_y_anchor): <type 'numpy.ndarray'>
    output_y_anchor.shape: (6, 1, 2)
    output_y_anchor:
     [[[ 0.  1.]]
    
     [[ 2.  3.]]
    
     [[ 4.  5.]]
    
     [[ 6.  7.]]
    
     [[ 8.  9.]]
    
     [[10. 11.]]]
    ++--++--++--++--++--++--++--++--
    type(output_y_reshape): <type 'numpy.ndarray'>
    output_y_reshape.shape: (6, 2)
    output_y_reshape:
     [[ 0.  1.]
     [ 2.  3.]
     [ 4.  5.]
     [ 6.  7.]
     [ 8.  9.]
     [10. 11.]]
    ++--++--++--++--++--++--++--++--
    type(output_y_split): <type 'list'>
    output_y_split:
     [array([[0., 1.]], dtype=float32), array([[2., 3.]], dtype=float32), array([[4., 5.]], dtype=float32), array([[6., 7.]], dtype=float32), array([[8., 9.]], dtype=float32), array([[10., 11.]], dtype=float32)]
    ++--++--++--++--++--++--++--++--
    output_y_split[0-5]:
    type(output_y_split[0]):<type 'numpy.ndarray'>
    output_y_split[0]:
     [[0. 1.]]
    type(output_y_split[1]):<type 'numpy.ndarray'>
    output_y_split[1]:
     [[2. 3.]]
    type(output_y_split[2]):<type 'numpy.ndarray'>
    output_y_split[2]:
     [[4. 5.]]
    type(output_y_split[3]):<type 'numpy.ndarray'>
    output_y_split[3]:
     [[6. 7.]]
    type(output_y_split[4]):<type 'numpy.ndarray'>
    output_y_split[4]:
     [[8. 9.]]
    type(output_y_split[5]):<type 'numpy.ndarray'>
    output_y_split[5]:
     [[10. 11.]]
    ++--++--++--++--++--++--++--++--
    output_y_split[0-5]:
    type([output_y_split[0]]):<type 'list'>
    [output_y_split[0]]:
     [array([[0., 1.]], dtype=float32)]
    type([output_y_split[1]]):<type 'list'>
    [output_y_split[1]]:
     [array([[2., 3.]], dtype=float32)]
    type([output_y_split[2]]):<type 'list'>
    [output_y_split[2]]:
     [array([[4., 5.]], dtype=float32)]
    type([output_y_split[3]]):<type 'list'>
    [output_y_split[3]]:
     [array([[6., 7.]], dtype=float32)]
    type([output_y_split[4]]):<type 'list'>
    [output_y_split[4]]:
     [array([[8., 9.]], dtype=float32)]
    type([output_y_split[5]]):<type 'list'>
    [output_y_split[5]]:
     [array([[10., 11.]], dtype=float32)]
    ++--++--++--++--++--++--++--++--
    
    Process finished with exit code 0

## 7. Example ##

#!/usr/bin/env python
    # -*- coding: utf-8 -*-
    
    from __future__ import absolute_import
    from __future__ import print_function
    from __future__ import division
    
    import os
    import sys
    import numpy as np
    import tensorflow as tf
    
    sys.path.append(os.path.dirname(os.path.abspath(__file__)))
    current_directory = os.path.dirname(os.path.abspath(__file__))
    
    print(16 * "++--")
    print("current_directory:", current_directory)
    print(16 * "++--")
    
    batch_size = 2
    num_step = 6
    num_input = 2
    
    # x_anchor shape: (batch_size, num_step, num_input)
    x_anchor = tf.constant([[[0, 1],
                             [2, 3],
                             [4, 5],
                             [6, 7],
                             [8, 9],
                             [10, 11]],
                            [[12, 13],
                             [14, 15],
                             [16, 17],
                             [18, 19],
                             [20, 21],
                             [22, 23]]], dtype=np.float32)
    
    # Permute num_step and batch_size.
    y_anchor = tf.transpose(x_anchor, perm=[1, 0, 2])
    
    # (num_step * batch_size, num_input)
    y_reshape = tf.reshape(y_anchor, [num_step * batch_size, num_input])
    
    # Split data because rnn cell needs a list of inputs for the RNN inner loop num_step * (batch_size, num_input)
    y_split = tf.split(y_reshape, num_step, 0)
    
    with tf.Session() as sess:
        input_x_anchor = sess.run(x_anchor)
        print("type(input_x_anchor):", type(input_x_anchor))
        print("input_x_anchor.shape:", input_x_anchor.shape)
        print(8 * "++--")
    
        output_y_anchor = sess.run(y_anchor)
        print("type(output_y_anchor):", type(output_y_anchor))
        print("output_y_anchor.shape:", output_y_anchor.shape)
        print("output_y_anchor:\n", output_y_anchor)
        print(8 * "++--")
    
        output_y_reshape = sess.run(y_reshape)
        print("type(output_y_reshape):", type(output_y_reshape))
        print("output_y_reshape.shape:", output_y_reshape.shape)
        print("output_y_reshape:\n", output_y_reshape)
        print(8 * "++--")
    
        output_y_split = sess.run(y_split)
        print("type(output_y_split):", type(output_y_split))
        print("output_y_split:\n", output_y_split)
        print(8 * "++--")
    
        print("output_y_split[0-5]:")
        for step in range(num_step):
            print("type(output_y_split[%d]):%s" % (step, type(output_y_split[step])))
            print("output_y_split[%d]:\n" % (step), output_y_split[step])
        print(8 * "++--")
    
        print("output_y_split[0-5]:")
        for step in range(num_step):
            print("type([output_y_split[%d]]):%s" % (step, type([output_y_split[step]])))
            print("[output_y_split[%d]]:\n" % (step), [output_y_split[step]])
        print(8 * "++--")

/usr/bin/python2.7 /home/strong/tensorflow_work/R2CNN_Faster-RCNN_Tensorflow/yongqiang.py
    ++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--
    current_directory: /home/strong/tensorflow_work/R2CNN_Faster-RCNN_Tensorflow
    ++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--++--
    2019-08-19 11:18:03.818784: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
    2019-08-19 11:18:03.896519: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:892] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
    2019-08-19 11:18:03.896755: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1030] Found device 0 with properties: 
    name: GeForce GTX 1080 major: 6 minor: 1 memoryClockRate(GHz): 1.7335
    pciBusID: 0000:01:00.0
    totalMemory: 7.92GiB freeMemory: 7.38GiB
    2019-08-19 11:18:03.896766: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1120] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: GeForce GTX 1080, pci bus id: 0000:01:00.0, compute capability: 6.1)
    type(input_x_anchor): <type 'numpy.ndarray'>
    input_x_anchor.shape: (2, 6, 2)
    ++--++--++--++--++--++--++--++--
    type(output_y_anchor): <type 'numpy.ndarray'>
    output_y_anchor.shape: (6, 2, 2)
    output_y_anchor:
     [[[ 0.  1.]
      [12. 13.]]
    
     [[ 2.  3.]
      [14. 15.]]
    
     [[ 4.  5.]
      [16. 17.]]
    
     [[ 6.  7.]
      [18. 19.]]
    
     [[ 8.  9.]
      [20. 21.]]
    
     [[10. 11.]
      [22. 23.]]]
    ++--++--++--++--++--++--++--++--
    type(output_y_reshape): <type 'numpy.ndarray'>
    output_y_reshape.shape: (12, 2)
    output_y_reshape:
     [[ 0.  1.]
     [12. 13.]
     [ 2.  3.]
     [14. 15.]
     [ 4.  5.]
     [16. 17.]
     [ 6.  7.]
     [18. 19.]
     [ 8.  9.]
     [20. 21.]
     [10. 11.]
     [22. 23.]]
    ++--++--++--++--++--++--++--++--
    type(output_y_split): <type 'list'>
    output_y_split:
     [array([[ 0.,  1.],
           [12., 13.]], dtype=float32), array([[ 2.,  3.],
           [14., 15.]], dtype=float32), array([[ 4.,  5.],
           [16., 17.]], dtype=float32), array([[ 6.,  7.],
           [18., 19.]], dtype=float32), array([[ 8.,  9.],
           [20., 21.]], dtype=float32), array([[10., 11.],
           [22., 23.]], dtype=float32)]
    ++--++--++--++--++--++--++--++--
    output_y_split[0-5]:
    type(output_y_split[0]):<type 'numpy.ndarray'>
    output_y_split[0]:
     [[ 0.  1.]
     [12. 13.]]
    type(output_y_split[1]):<type 'numpy.ndarray'>
    output_y_split[1]:
     [[ 2.  3.]
     [14. 15.]]
    type(output_y_split[2]):<type 'numpy.ndarray'>
    output_y_split[2]:
     [[ 4.  5.]
     [16. 17.]]
    type(output_y_split[3]):<type 'numpy.ndarray'>
    output_y_split[3]:
     [[ 6.  7.]
     [18. 19.]]
    type(output_y_split[4]):<type 'numpy.ndarray'>
    output_y_split[4]:
     [[ 8.  9.]
     [20. 21.]]
    type(output_y_split[5]):<type 'numpy.ndarray'>
    output_y_split[5]:
     [[10. 11.]
     [22. 23.]]
    ++--++--++--++--++--++--++--++--
    output_y_split[0-5]:
    type([output_y_split[0]]):<type 'list'>
    [output_y_split[0]]:
     [array([[ 0.,  1.],
           [12., 13.]], dtype=float32)]
    type([output_y_split[1]]):<type 'list'>
    [output_y_split[1]]:
     [array([[ 2.,  3.],
           [14., 15.]], dtype=float32)]
    type([output_y_split[2]]):<type 'list'>
    [output_y_split[2]]:
     [array([[ 4.,  5.],
           [16., 17.]], dtype=float32)]
    type([output_y_split[3]]):<type 'list'>
    [output_y_split[3]]:
     [array([[ 6.,  7.],
           [18., 19.]], dtype=float32)]
    type([output_y_split[4]]):<type 'list'>
    [output_y_split[4]]:
     [array([[ 8.,  9.],
           [20., 21.]], dtype=float32)]
    type([output_y_split[5]]):<type 'list'>
    [output_y_split[5]]:
     [array([[10., 11.],
           [22., 23.]], dtype=float32)]
    ++--++--++--++--++--++--++--++--
    
    Process finished with exit code 0

[https_github.com_tensorflow_docs_blob_r1.3_site_en_api_docs_api_docs_python_tf_split.md]: https://github.com/tensorflow/docs/blob/r1.3/site/en/api_docs/api_docs/python/tf/split.md