jizhihang / awf-matrix-factorization

A simple implementation of low-rank matrix factorization using MATLAB's built-in Levenberg Marquardt.

Matrix Factorization

A simple implementation of low-rank matrix factorization using MATLAB's built-in Levenberg Marquardt (or the simple LM from awful.codeplex.com).

Usage:

%1. Obtain M, W, e.g.

    [M,W] = load_giraffe;

%2. Choose initial estimates A0, B0

    A0 = rand(size(M,1), 6);
    B0 = rand(size(M,2), 6);

%3. Call

    opts = awf_mf_lsqnonlin('opts');
    opts.lsopts.Display = 'iter';
    [A,B] = awf_mf_lsqnonlin(W,M,A0,B0,opts);
    tic
    opts.lsopts.Display = 'none';
    [A,B] = awf_mf_lsqnonlin(W,M,A0,B0,opts);
    toc % will take about 1 minute 

%4. or, with prior and/or Gauge fixing

	opts.regularizer_lambda = 1e-5;
	opts.gauge_fix_weight = 1e-5;
	[A,B] = awf_mf_lsqnonlin(W,M,A0,B0,opts);

%5. or, with au_levmarq

    opts = awf_mf_lsqnonlin('opts');
	opts.Algorithm = 'awf';
    tic
	[A,B] = awf_mf_lsqnonlin(W,M,A0,B0,opts);
    toc

Sources

damped_wiberg.zip from [http://www.vision.is.tohoku.ac.jp/us/download] md5 hash 6E72535A32F06045C9C7EC626A38D543

Data from [http://www.robots.ox.ac.uk/~amb]