5 Tensor Operations for Linear Algebra and Data Processing with PyTorch

A short introduction about PyTorch and about the following functions:

torch.roll(input, shifts, dims);
torch.svd(input, some=True, compute_uv=True, out=None);
torch.eig(eigenvectors=False);
torch.nonzero(input, out=None, as_tuple=False);
torch.norm(input, p='fro', dim=None, keepdim=False, out=None, dtype=None).

# Import torch
import torch

# Creating 2D tensors for further processing
mat = []

print('Original matrices:')
for i in range(4):
    m = (i+1) * torch.ones(3,3, dtype=torch.int32)
    mat.append(m)
    print("Matrix {}:\n{}\n".format(i+1,m.numpy()))

Original matrices:
Matrix 1:
[[1 1 1]
 [1 1 1]
 [1 1 1]]

Matrix 2:
[[2 2 2]
 [2 2 2]
 [2 2 2]]

Matrix 3:
[[3 3 3]
 [3 3 3]
 [3 3 3]]

Matrix 4:
[[4 4 4]
 [4 4 4]
 [4 4 4]]

Function 1 - roll(shifts, dims) → Tensor

Rolls a tensor along the chosen dimensions per specified amount of shifts.

# Example 1 - preparing original tensor:
x = torch.ones(3,3)
X = torch.stack((x, 2*x), dim=0)
X

tensor([[[1., 1., 1.],
         [1., 1., 1.],
         [1., 1., 1.]],

        [[2., 2., 2.],
         [2., 2., 2.],
         [2., 2., 2.]]])