function mapping = cluster_to_class( sz_clusters, class_idx, cluster_idx, varargin )
%CLUSTER_TO_CLASS Maps clusters to classes
global g_config;
addpath(fullfile(fileparts(mfilename('fullpath')), './extern/'));
[min_samples_p, min_samples_exp, discard_mixed] = process_options(varargin, ...
'MinSamplesPercentage', g_config.CLUSTER_CLASS_MINIMUM_SAMPLES_P, ...
'MinSamplesExponent', g_config.CLUSTER_CLASS_MINIMUM_SAMPLES_EXP, ...
'DiscardMixed', 1);
classes = sort(unique([class_idx{:}]));
nclasses = length(classes);
nclusters = length(sz_clusters);
mapping = zeros(1, nclusters);
for j = 1:nclusters
% count elements of each class
nmax = 0;
iclass = 0;
for k = 1:nclasses
n = 0;
% count elements of this class
for i = 1:length(cluster_idx)
if cluster_idx(i) == j % match cluster
if any(classes(k) == class_idx{i})
n = n + 1;
else
% see if we have other classes here
if discard_mixed && any(class_idx{i} > 0)
% other classes present, move to next class
n = 0;
break;
end
end
end
end
if n > nmax
nmax = n;
iclass = k;
end
end
% see if we have the minimum number of samples (proportional to the
% cluster size)
if sz_clusters(j) > 0
pmin = max(min_samples_p, (1/sz_clusters(j))^min_samples_exp);
if iclass ~= 0 && nmax >= pmin*sz_clusters(j)
mapping(j) = classes(iclass);
end
end
end
end