TensorFlow Stack and unstack detailed introduction __tensorflow

Turn from: https://www.toutiao.com/i6463423755296178702/

Tf.stack ()

Tf.stack (values, axis=0, name= ' stack ')
Converts a tensor array of one dimension R to a r+1 tensor with the specified axis axis. Increases a dimension by assigning a set of tensor to a specified axis.

Suppose the tensor array to be converted is of the length n, where the shape of each tensor is (A, B, C).
If the axis is axis=0, the shape of the transformed tensor is (N, A, B, C).
If the axis is Axis=1, the shape of the transformed tensor is (A, N, B, C).
If the axis is axis=2, the shape of the transformed tensor is (A, B, N, C). The other case is the analogy.

Examples are as follows:
' X ' is [1, 4], Shape is (2), dimension is 1
' Y ' is [2, 5], Shape is (2), dimension is 1
' Z ' is [3, 6], Shape is (2), dimension is 1
Stack ([x, Y, z]) => [[1, 4], [2, 5], [3, 6]] # Axis's value defaults to 0. The output shape is (3, 2)
Stack ([x, Y, z], Axis=1) => [[1, 2, 3], [4, 5, 6]] The value of axis is 1. The output shape is (2, 3)

' X ' is [[1,1,1,1],[2,2,2,2],[3,3,3,3]], the shape is (3,4), the dimension is 2
' Y ' is [[4,4,4,4],[5,5,5,5],[6,6,6,6]], the shape is (3,4), the dimension is 2
Stack ([x,y]) => [[[[1,1,1,1],[2,2,2,2],[3,3,3,3]]], [[4,4,4,4],[5,5,5,5],[6,6,6,6]]] # axis defaults to 0. The output shape is (2, 3, 4)
The value of stack ([X,y],axis=1) => [[[1,1,1,1],[4,4,4,4]],[[2,2,2,2],[5,5,5,5]],[[3,3,3,3],[6,6,6,6]]] # axis is 1. The output shape is (3, 2, 4)
Stack ([x,y],axis=2) => [[1,4],[1,4],[1,4],[1,4]],[[2,5],[2,5],[2,5],[2,5]],[[3,6],[3,6],[3,6],[3,6]]]# Axis has a value of 2. The output shape is (3, 4, 2)

Axis is able to understand that stack is a dimension that increases the tensor of a group of identical shapes. Axis is the tensor in which axis-specified dimensions are replaced with all of these tensor arrays. For example, the axis=2 is specified in the 2nd dimension, and the original element is replaced with the elements of the entire tensor array, from (A, B, c) to (A, B, N, C).

Parameters:
Values: An array of tensor arrays with the same shape and data type.
Axis: An integer that is centered around shaft axis. The default is the first dimension, axis=0. Supports negative numbers. Value range is [-(r+1), r+1)
Name: This operation is named (optional)
Returns: Tensor after being raised by one dimension
Exception: ValueError: If axis axis goes out of range [-(r+1), r+1). Tf.unstack ()

Tf.unstack (value, Num=none, axis=0, name= ' Unstack ')
Converts a tensor array of one dimension R to a R-1 tensor with the specified axis axis. Reduces a dimension by a set of tensor to a specified axis. Just as opposed to stack ().

Splits the tensor value into a number of num tensor arrays. If NUM is not specified, it is specified according to the shape of the tensor value. If Value.shape[axis] does not exist, the ValueError exception is thrown.

If the shape of a tensor is (A, B, C, D).
if Axis = = 0, the output tensor is value[i,:,:,:],i takes a value of [0,a), and the shape of the tensor for each output is (b,c,d).
if Axis = = 1, the output tensor is value[:, I,:,:],i takes a value of [0,b), and the shape of the tensor for each output is (a,c,d).
if Axis = = 2, the output tensor is value[:,:, I,:],i takes a value of [0,c), and the shape of the tensor for each output is (a,b,d). by analogy.

Examples are as follows:
' X ' is [[1,1,1,1],[2,2,2,2],[3,3,3,3]] # shape is (3,4) with a dimension of 2
Unstack (x,axis=0) => with the specified dimension 0 as an axis, to 3 shapes (4) tensor [1,1,1,1],[2,2,2,2],[3,3,3,3]
Unstack (X,axis=1) => with the specified dimension 1 as an axis, to 4 shapes (3) tensor [1,2,3],[1,2,3],[1,2,4],[1,2,3]

Axis can understand this: Unstack is an array of tensor that is to reduce one tensor to a lower dimension. Axis is the dimension specified by axis, which is replaced by the data in the same dimension as all of this tensor.

Parameter:
Value: A tensor that is greater than 0 for a dimension that will be reduced.
Num: Integer. The length of the specified dimension axis. If set to None (default), the value is automatically evaluated.
Axis: Integer. The dimension that is specified with Axis axis is the first dimension, axis=0, that is the default. Supports negative numbers. The value range is [R, R]
Name: The name of this action (optional)
Returns:
An array of tensor values that is reduced from the dimension of tensor value.
Exception:
ValueError: If num is not specified and cannot be obtained.
ValueError: If axis goes out of range [-R, R].