cosmo_anova_feature_selector.m

function selected_indices=cosmo_anova_feature_selector(dataset, ratio_to_keep)
% using an anove finds the features that show the most variance between
% classes
%
% selected_indices=cosmo_anova_feature_selector(dataset, ratio_to_keep)
%
% Inputs
%  dataset          struct with .samples and .sa.targets
%  ratio_to_keep    value between 0 and 1
%
% Output
%  selected_indices   feature ids in dataset with most variance between
%                     classes. len(selected_indices) is approximately 
%                     equal to ratio_to_keep * size(dataset.samples,2)
%                     
% NNO Aug 2013

targets=dataset.sa.targets;
samples=dataset.samples;
[nsamples, nfeatures]=size(samples);

ps=zeros(nfeatures,1);
for k=1:nfeatures
    ps(k)=anova1(samples(:,k), targets, 'off');
end

[foo, idxs]=sort(ps);
n_idxs=round(ratio_to_keep*nfeatures);

selected_indices=idxs(1:n_idxs);