PyTorch - Calculate one step of gradient

Calculate one step of gradient descent for the following tensor:

f(X)=∑log_e​(x_ij​+3)

X - tensor 3x3

X^t=0 = [[1,2,3],[5,6,7],[8,9,0]]

α=5 - gradient coefficient

find X^t=1

Briefly, it is necessary to find values after applying one step towards gradient.

import torch

# create initial tensor with floats
w = torch.tensor( [[1.,2.,3.],[5.,6.,7.],[8.,9.,0.]], requires_grad=True)

# or it is possible to specify type of data directly
#w = torch.tensor( [[1,2,3],[5,6,7],[8,9,0]], requires_grad=True, dtype=torch.float)

a = 5
function = a * torch.log(w + 3.).sum()

# another way to write functions for torch:
#function = a * (w + 3.).log().sum()

function.backward()
print(w-w.grad)

#tensor([[-0.2500,  1.0000,  2.1667],
#        [ 4.3750,  5.4444,  6.5000],
#        [ 7.5455,  8.5833, -1.6667]], grad_fn=)

Published: 2022-05-04 22:54:23
Updated: 2022-05-05 02:17:38

Last 10 artitles

• 2023-12-25 01:48:38 - In-place and out-place Numpy functions
• 2023-12-25 00:58:08 - First order ordinary differential equations
• 2023-12-25 00:22:15 - How to increase swap space in Ubuntu
• 2023-12-24 04:03:12 - How to accelerate math calculations with Numba and Cython
• 2023-12-19 00:38:20 - 3 ways of function caching in Python
• 2023-12-19 00:16:55 - Accelerating Python
• 2023-12-16 12:26:55 - BioPython
• 2023-12-02 11:05:34 - Reading FITS files with astropy
• 2023-12-02 05:23:42 - Astropy
• 2023-12-01 00:33:16 - Circle length calculation with the Monte-Carlo method

9 popular artitles

• Python Panda (3443)
• Python NumPy (2979)
• Python DICTionary HowTo (1591)
• Python LIST HowTo (1471)
• Python (1422)
• Python Generators (1281)
• Python: How to make absolute links in BeautifulSoup (1084)
• Programming language (967)
• NumPy: What the difference between mgrid and meshgrid (946)