Today, the code used tf.repeat()

I checked the information and no one specifically explained this. Let me write it, hoping to help future generations.

Official documents:

https://www.tensorflow.org/api_docs/python/tf/repeat?hl=ca

Call the method:

tf.repeat(input, repeats, axis=None, name=None)

parameter:
1) input: a tensor

2) repeats: the number of repetitions
Note: len(repeats) must equal input.shape[axis] if axis is not None

3) axis: dimension
Let’s see an example to understand
Description: An int. The axis along which to repeat values. By default (axis=None), use the flattened input array, and return a flat output array.
If the axis has no parameters, the array will be flattened first to become one-dimensional and then repeated

Example 3:

tf.repeat([[1, 2], [3, 4]], repeats=[2, 3], axis=1)
                                    -Input the 0th element repeated 2 times
                                                                         -Input the first element to repeat 3 times
[
 [1, 1, 2, 2, 2],
 [3, 3, 4, 4, 4]
]

tf.repeat([[1, 2], [3, 4]], repeats=[2, 3], axis=0)
                                                                         -Input the 0th element repeated 2 times
                                                                         -Input the first element to repeat 3 times
[
 [1, 2],
 [1, 2],
 [3, 4],
 [3, 4],
 [3, 4]
]

For simple memory, axis = 1 increases horizontally, axis = 0 increases vertically

Example 1:
First look at the simplest one-dimensional

temp = tf.constant([1,2,3])
tf.Tensor: shape=(1, 3), dtype=int32, numpy=array([[1, 2, 3]], dtype=int32)

tf.repeat(input = temp, repeats=3, axis = 0)
 Output:
[1, 1, 1, 2, 2, 2, 3, 3, 3]
 Each number repeated 3 times

When the input is not one-dimensional, but no axis is assigned, the input will be flattened into one-dimensional

temp1 = tf.constant([[1,2],[3,4]])
tf.repeat(input = temp1, repeats = 2)
 Output:
[1,1,2,2,3,3,4,4]

Example 2:

temp3 = tf.const([[1],[2],[3]])
temp3.shape ---> (3,1)

As mentioned above, len(repeats) = input.shape[axis]

So in this example it must satisfy:
temp3.shape[axis = 0] = 4 = len(repeats)
temp3.shape[axis = 1] = 1 = len(repeats)

 1) When axis = 0
tf.repeat(temp3, repeats = [1,2,3], axis = 0)
 Output:
[
[1],
[2],
[2],
[3],
[3],
[3],
]

 2) When axis = 1
tf.repeat(temp3, repeats=2, axis=1)
 Output:
[
 [1, 1],
 [2, 2],
 [3, 3],
 [4, 4]
]