netflix_svd.m

%% Read Netflix dataset
A = readSMAT('/scratch/dgleich/netflix/netflix.smat');

k = [10 25 50 100 150 200];
l = size(k,2);

%% Matlab's SVDS
for i= 1:l
    tic; 
    [U,S,V] = svds(A,k(i)); 
    toc
    file = strcat('test1_',int2str(k(i)), '.mat');
    d = diag(S);
    save(file, 'd');
end

%%  Using eigs() on A*A'
m = size(A,1);
Cx = @(x) A*(A'*x);

for i= 1:l
    tic;
    [V D]=eigs(Cx,m,k(i),'LA',struct('issym',1,'disp',0));
    toc
    file = strcat('test2_',int2str(k(i)), '.mat');
    d = diag(D);
    save(file, 'd');
end

%%  Using Propack.
addpath('/homes/hou13/fall2013/MR_SVD_test/PROPACK/matlab');
m = size(A,1); n = size(A,2);

Ax=@(x) A*x;
Atx=@(x) A'*x;

for i= 1:l
    tic;
    [UD D VD]=lansvd(Ax,Atx,m,n,k(i),'L');
    toc
    file = strcat('test3_',int2str(k(i)), '.mat');
    d = diag(D);
    save(file, 'd');
end

netflix_svd.py

import numpy as np
from time import time

from scipy.sparse import csc_matrix
from scipy.linalg import svd
from pypropack import svdp
from scipy.sparse.linalg import svds
from sparsesvd import sparsesvd

import scipy.io

if __name__ == '__main__':
	D = scipy.io.loadmat('/u/puma/hou13/fall2013/MR_SVD_test/netflix.mat')
	M = D['A']
	#M = '/scratch/dgleich/netflix/netflix.smat'
	ks = [10, 25, 50, 100, 150, 200]
	# PROPACK
	times1 = []
	for k in ks:
		t0 = time()
        res1 = svdp(M, k, kmax = 1000, compute_u=False, compute_v=False)
        t1 = time()
        t = t1 - t0
        times1.append(t)
        print k, t
    scipy.io.savemat('test1.mat',{'d':res1})
        
        
    # ARPACK
	times2 = []
	for k in ks:
		t0 = time()
        res2 = svds(M, k, return_singular_vectors = False)
        t1 = time()
        t = t1 - t0
        times2.append(t)
        print k, t
    scipy.io.savemat('test2.mat',{'d':res2})
        
    # SVDLIBC
	times3 = []
	for k in ks:
		t0 = time()
        u,res3,v = sparsesvd(M, k)
        t1 = time()
        t = t1 - t0
        times1.append(t)
        print k, t
    scipy.io.savemat('test3.mat',{'d':res3})