Transpose¶

Description¶

Info

Parent class: Module

Derived classes: -

This module implements a permutation of the input tensor axes in the desired order. An analog of the transpose function in numpy.

def __init__(self, axes=None, name=None):

Parameters

Parameter	Allowed types	Description	Default
axes	tuple	Tuple with the desired axis layout	None
name	str	Layer name	None

Explanations

axes - if None, then the usual transpose operation will be performed, otherwise, the permutation is performed in the specified order.

Necessary imports.

import numpy as np
from PuzzleLib.Backend import gpuarray
from PuzzleLib.Modules import Transpose

Info

gpuarray is required to properly place the tensor in the GPU

shape = (10, 3, 5, 4, 2)
axes = (2, 4, 1, 3, 0)
data = gpuarray.to_gpu(np.random.randn(*shape).astype(np.float32))

Let us initialize the operation with the default parameters (axes=None):

transpose = Transpose()
outdata = transpose(data)
print(outdata.shape)

(2, 4, 5, 3, 10)

This time let us set the desired axis order:

transpose = Transpose(axes)
outdata = transpose(data)
print(outdata.shape)

(5, 2, 3, 4, 10)