PyTorch torch.linalg.pinv Function


Pytorch torch 参考手册Pytorch torch Reference Manual

torch.linalg.pinvIt is a function in the PyTorch linear algebra module used to compute the Moore-Penrose pseudo-inverse of a matrix. For non-square or singular matrices, the pseudo-inverse is an important tool.

Function Definition

torch.linalg.pinv(A, rcond=None, hermitian=False, out=None)

Parameters:

  • A(Tensor): The input matrix.
  • rcond(float, optional): Singular value truncation threshold.
  • hermitian(bool, optional): If True, the matrix is assumed to be Hermitian. Defaults to False.
  • out(Tensor, optional): The output tensor.

Return Value:

  • torch.Tensor: Returns the pseudo-inverse of the matrix.

Usage Example

Example

import torch

# Create a non-square matrix
A = torch.tensor([[1.0, 2.0, 3.0],
                  [4.0, 5.0, 6.0]], dtype=torch.float64)

# Compute the pseudo-inverse
A_pinv = torch.linalg.pinv(A)

print("Matrix A:")
print(A)
print("nPseudo-inverse A^+:")
print(A_pinv)
print("nVerification: A @ A^+ @ A ≈ A")
print(A @ A_pinv @ A)

The output result is:

矩阵 A:
tensor([[1., 2., 3.],
        [4., 5., 6.]], dtype=torch.float64)
伪逆 A^+:
tensor([[-0.9444,  0.4444],
        [-0.2778,  0.2778],
        [ 0.3889,  0.1111]], dtype=float64)
验证: A @ A^+ @ A ≈ A
tensor([[ 1.0000e+00,  2.0000e+00,  3.0000e+00],
        [ 4.0000e+00,  5.0000e+00,  6.0000e+00]], dtype=float64)

Pytorch torch 参考手册Pytorch torch Reference Manual

Other Extensions