STA-663-Final-Project-SSVD 0.1.2

Optimized Sparse Singular Value Decomposition with Numba

STA-663-Final-Project

Authors: Jingxuan Zhang, Jae Hyun Lee

This is package for final project in STA-663 Python programming course.

This package can be used by install STA-663-Final-Project-SSVD from pypi

In this package, ssvd_opt function is included which is implementation of paper Biclustering via Sparse Singular Value Decomposition by Lee et al.

ssvd_opt finds latent association between columns and rows under High Dimension Low Sample Size. This function is optimized with functions from Numba. Therefore, before execution, you should import njit, prange, jit from Numba.

ssvd_opt

input: X, niter

X - data matrix which has much more columns than rows

niter - Number of max iteration

output: u,s,v,iters

u,s,v - Decomposed singular value and singular vectors with adaptive lasso penalty which corresponds to best rank 1 approximation matrix X* = s*u@v.T.

iters - number of iteration before convergence.

if the algorithm fails to converge, it prints out "need to increase niter!"