function test_suite = test_distatis
% tests for cosmo_distatis
%
% # For CoSMoMVPA's copyright information and license terms, #
% # see the COPYING file distributed with CoSMoMVPA. #
try % assignment of 'localfunctions' is necessary in Matlab >= 2016
test_functions = localfunctions();
catch % no problem; early Matlab versions can use initTestSuite fine
end
initTestSuite;
function test_statis_
% using: 1. Abdi, H. & Valentin, D. in Encyclopedia of Measurement
% and Statistics (Salkind, N.) 42?42 (SAGE Publications, 2007).
d = cell(0);
% note: element [1,3] is reported as .148, for symmetry use .146
d{1} = [0 0.1120 0.1460 0.0830 0.1860 0.1100
0.1120 0 0.1520 0.0980 0.1580 0.1340
0.1460 0.1520 0 0.2020 0.2850 0.2490
0.0830 0.0980 0.2020 0 0.1310 0.1100
0.1860 0.1580 0.2850 0.1310 0 0.1550
0.1100 0.1340 0.2490 0.1100 0.1550 0];
d{2} = [0 0.6000 1.9800 0.4200 0.1400 0.5800
0.6000 0 2.1000 0.7800 0.4200 1.3400
1.9800 2.1000 0 2.0200 1.7200 2.0600
0.4200 0.7800 2.0200 0 0.5000 0.8800
0.1400 0.4200 1.7200 0.5000 0 0.3000
0.5800 1.3400 2.0600 0.8800 0.3000 0];
d{3} = [0 0.5400 1.3900 5.7800 10.2800 6.7700
0.5400 0 1.0600 3.8000 6.8300 4.7100
1.3900 1.0600 0 8.0100 11.0300 5.7200
5.7800 3.8000 8.0100 0 2.5800 6.0900
10.2800 6.8300 11.0300 2.5800 0 3.5300
6.7700 4.7100 5.7200 6.0900 3.5300 0];
d{4} = [0 0.0140 0.1590 0.0040 0.0010 0.0020
0.0140 0 0.0180 0.0530 0.0240 0.0040
0.1590 0.0180 0 0.2710 0.0670 0.0530
0.0040 0.0530 0.2710 0 0.0010 0.0080
0.0010 0.0240 0.0670 0.0010 0 0.0070
0.0020 0.0040 0.0530 0.0080 0.0070 0];
ds = get_distance_dataset(d);
ds = cosmo_stack({ds, ds, ds}, 2); % features
cosmo_check_dataset(ds);
opt = struct();
opt.split_by = 'subject';
opt.return = 'crossproduct';
opt.progress = false;
res = cosmo_distatis(ds, opt);
% note: S_{[+]}[6,2] is reported as -0.01, should be -.100
s = [.176 .004 -.058 .014 -.100 -.036
.004 .178 .022 -.038 -.068 -.100
-.058 .022 .579 -.243 -.186 -.115
.014 -.038 -.243 .240 .054 -.027
-.100 -.068 -.186 .054 .266 .034
-.036 -.100 -.115 -.027 .034 .243];
u = cosmo_unflatten(res, 1);
assertElementsAlmostEqual(repmat(s, [1, 1, 3]), u, 'absolute', .001);
assertElementsAlmostEqual(res.fa.quality(2), .6551, 'absolute', .001);
opt.return = 'distance';
opt.shape = 'square';
res = cosmo_distatis(ds, opt);
u = cosmo_unflatten(res, 1);
sq = cosmo_squareform(u(:, :, 1));
assertElementsAlmostEqual(sq, [0.3452 0.8710 0.3888 0.6419 ...
0.4911 0.7112 0.4919 0.5789 ...
0.6203 1.3049 1.2163 1.0512 ...
0.3996 0.5354 0.4400], 'absolute', .001);
opt.shape = 'triangle';
resvec = cosmo_distatis(ds, opt);
assertElementsAlmostEqual(resvec.samples(:, 1), sq');
% test numeric input
vec_samples = mat2cell(ds.samples(:, 1), ones(4, 1) * 15, 1);
resvec2 = cosmo_distatis(vec_samples, opt);
assertElementsAlmostEqual(resvec2.samples, resvec.samples(:, 1));
assertEqual(cellfun(@numel, resvec2.a.sdim.values), ...
cellfun(@numel, resvec.a.sdim.values));
resvec2.a.sdim.values = resvec.a.sdim.values;
resvec_single_feature = cosmo_slice(resvec, 1, 2);
assertElementsAlmostEqual(resvec_single_feature.samples, ...
resvec2.samples);
assertElementsAlmostEqual(resvec_single_feature.fa.quality, ...
resvec2.fa.quality);
resvec2.samples = resvec_single_feature.samples;
resvec2.fa.quality = resvec_single_feature.fa.quality;
assertEqual(resvec2, resvec_single_feature);
opt.weights = 'uniform';
resvec = cosmo_distatis(ds, opt);
assertElementsAlmostEqual(resvec.samples(:, 1)', ...
[0.3156 0.8625 0.3704 0.6043 0.4646 ...
0.6572 0.4788 0.5489 0.5783 1.3221 ...
1.1576 0.9885 0.3644 0.5068 0.4037], ...
'absolute', .001);
% test exceptions
aet = @(varargin)assertExceptionThrown(@() ...
cosmo_distatis(varargin{:}, 'progress', false), '');
% no compromise possible
d2 = d;
d2{4}(1, 5) = 3;
d2{4}(5, 1) = 3;
opt.weights = 'eig';
ds2 = get_distance_dataset(d2);
aet(ds2, opt);
% illegal arguments
ds.sa.chunks = ds.sa.subject;
opt = struct();
opt.shape = 'foo';
aet(ds, opt);
opt = struct();
opt.weights = 'foo';
aet(ds, opt);
opt = struct();
opt.return = 'foo';
aet(ds, opt);
opt = struct();
% cannot deal with empty input
aet({}, opt);
% needs dataset or numeric input
aet({false, true}, opt);
% cannot take non-matrix input
aet({zeros([2 2 2])}, opt);
% illegal field in dataset
ds_bad = ds;
ds_bad.foo = 2;
opt = struct();
aet(ds_bad, opt);
function ds = get_distance_dataset(d)
nsubj = numel(d);
ds_all = cell(nsubj, 1);
for k = 1:nsubj
ds = struct();
sq = cosmo_squareform(d{k});
ds.samples = sq(:);
nd = size(d{k}, 1);
ns = size(ds.samples, 1);
ds.sa.subject = k * ones(ns, 1);
[i, j] = find(triu(repmat(1:nd, nd, 1)', 1)');
ds.sa.targets1 = i;
ds.sa.targets2 = j;
faces = {'f1', 'f2', 'f3', 'f4', 'f5', 'f6'}';
ds.a.sdim.values = {faces, faces};
ds.a.sdim.labels = {'targets1', 'targets2'};
ds_all{k} = ds;
end
ds = cosmo_stack(ds_all);
