Singular Value Decomposition (svd) in SciPy

Introduction

Singular Value Decomposition (SVD) breaks a matrix into simpler parts. It helps us understand data patterns and reduce complexity.

To compress images by keeping only important features.

To find hidden topics in text data.

To reduce the number of features in a dataset for easier analysis.

To solve systems of linear equations in a stable way.

To analyze relationships in recommendation systems.

Syntax

SciPy

from scipy.linalg import svd
U, s, VT = svd(A, full_matrices=True, lapack_driver='gesdd')

A is the input matrix you want to decompose.

U and VT are orthogonal matrices, s contains singular values.

Examples

Basic SVD on a 2x2 matrix.

SciPy

from scipy.linalg import svd
A = [[1, 2], [3, 4]]
U, s, VT = svd(A)

Use full_matrices=False to get smaller U and VT matrices.

SciPy

U, s, VT = svd(A, full_matrices=False)

Choose a different LAPACK driver for SVD calculation.

SciPy

U, s, VT = svd(A, lapack_driver='gesvd')

Sample Program

This code decomposes matrix A into U, s, and VT using SVD. It prints each part so you can see the result.

SciPy

from scipy.linalg import svd
import numpy as np

# Create a 3x2 matrix
A = np.array([[1, 2], [3, 4], [5, 6]])

# Perform SVD
U, s, VT = svd(A)

print('Matrix U:')
print(U)
print('\nSingular values:')
print(s)
print('\nMatrix VT:')
print(VT)

Important Notes

The singular values in s are always sorted from largest to smallest.

You can reconstruct the original matrix by multiplying U * diag(s) * VT.

Using full_matrices=False can save memory for large matrices.

Summary

SVD splits a matrix into three parts: U, singular values, and VT.

It helps find important patterns and reduce data size.

Scipy's svd function is easy to use and flexible.