Example of de-meaning

Contents

Generate random dataset

ds = cosmo_synthetic_dataset('nchunks', 4, 'ntargets', 3);

% add some constant to all data
ds.samples = ds.samples + 2;

% show dataset
subplot(2, 2, 1);
imagesc(ds.samples, [-4 4]);
title('before demeaning');
subplot(2, 2, 2);
hist(ds.samples(:), 10);
xlim([-6 6]);

Split the dataset by chunks

>@@>

splits = cosmo_split(ds, {'chunks'}, 1);
% <@@<
nsplits = numel(splits);

% allocate space for output
outputs = cell(nsplits, 1);

% treat each element in splits separately, and subtract the mean for each
% feature separately
for k = 1:nsplits
    d = splits{k};
    % >@@>

    % mean over samples, for each feature
    mu = mean(d.samples, 1);

    % subtract the mean.
    % equivalent, but less efficient, is:
    %     nsamples=size(d.samples,1);
    %     d.samples=d.samples-repmat(mu,nsamples,1);
    %
    d.samples = bsxfun(@minus, d.samples, mu);
    % <@@<

    % store output
    outputs{k} = d;
end

ds_demeaned = cosmo_stack(outputs);

% show dataset
subplot(2, 2, 3);
imagesc(ds_demeaned.samples, [-4 4]);
title('after demeaning');
subplot(2, 2, 4);
hist(ds_demeaned.samples(:), 10);
xlim([-6 6]);

Alternative approach to demeaning

% note: the samples in the output are in a different order than the input,
% but otherwise the same
demeaner = @(x)bsxfun(@minus, x, mean(x, 1)); % function handle as helper
ds_demeaned_alt = cosmo_fx(ds, demeaner, 'chunks');

Published with MATLAB® R2024a