Added plain Nystrom KRLS algorithm with Tikhonov regularization. It is useful for instance for datasets in which the number of training points n is so large that the kernel matrix K does not fit in memory.
See: http://arxiv.org/abs/1303.1849
Cross validation of sigma and lambda supported.
Only supports Gaussian kernel at the moment.
Example:
load(fullfile(gurls_root,'demo/data/breastcancer_data.mat'));
% Gaussian kernel approximated with Nystrom subsampling (m=200).
% Hold Out cross validation to select lambda and the Kernel width sigma
name = 'nysrbfho';
opt = gurls_defopt(name);
opt.seq = {'split:ho', 'paramsel:siglamho_nystrom', 'kernel:rbf_nystrom', 'rls:dual_nystrom', 'predkernel:traintest_nystrom', 'pred:dual_nystrom', 'perf:rmse'};
opt.process{1} = [2,2,2,2,0,0,0];
opt.process{2} = [3,3,3,3,2,2,2];
opt.nystrom.shuffle = 0;
opt.nystrom.m = 200;
opt.hoperf = @perf_rmse;
gurls (Xtr, ytr, opt,1);
gurls (Xte, yte, opt,2);
Added plain Nystrom KRLS algorithm with Tikhonov regularization. It is useful for instance for datasets in which the number of training points n is so large that the kernel matrix K does not fit in memory. See: http://arxiv.org/abs/1303.1849 Cross validation of sigma and lambda supported. Only supports Gaussian kernel at the moment.
Example: