<!DOCTYPE html>
<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
</head>
<body>
<p><font size="2">Hi, <br>
</font></p>
<p><font size="2">I would like to use the FT permutations test with
cluster based correction on 2D data freq x time with a within
subject design. Data are obtained after sources reconstruction
and are for example related to a particular ROI (e.g. V5_R). The
first dimension is a frequency dimension but not the oscillatory
brain frequency. This dimension is linked to a temporal
frequency of the input visual stimulus. the second dimension is
time (sRate = 1kHz) and data are timelocked ERP for different
values of this temporale frequency (by step of 1Hz). The data
therefore has the same structure as if it had been obtained by
time-frequency analysis on a particular electrode. <br>
</font></p>
<p><font size="2">For example, here's the data for one participant:<br>
ftAllFiles{1}<br>
<br>
ans = <br>
<br>
struct with fields:<br>
<br>
dimord: 'freq_time'<br>
avg: [29×401 double]<br>
time: [-0.2000 -0.1990 -0.1980 -0.1970 -0.1960 -0.1950
-0.1940 -0.1930 -0.1920 -0.1910 -0.1900 -0.1890 -0.1880 -0.1870
-0.1860 … ]<br>
freq: [6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
24 25 26 27 28 29 30 31 32 33 34]<br>
label: {'V5_R'}<br>
There are 29 participants for each conditions. ftAllFiles have
2x29 cells, the first 29 correspond to the first condition and
the last 29 to the second condition. <br>
</font></p>
<p><font size="2">Here is the configuration : <br>
</font></p>
<p><font size="2">>> statcfg<br>
<br>
statcfg = <br>
<br>
struct with fields:<br>
<br>
frequency: 'all'<br>
parameter: 'avg'<br>
avgoverfreq: 'no'<br>
avgoverchan: 'no'<br>
latency: 'all'<br>
method: 'montecarlo'<br>
statistic: 'ft_statfun_depsamplesT'<br>
correctm: 'cluster'<br>
clusteralpha: 0.0500<br>
clusterstatistic: 'maxsum'<br>
minnbchan: 0<br>
tail: 1<br>
clustertail: 1<br>
alpha: 0.0500<br>
numrandomization: 100000<br>
ivar: 1<br>
uvar: 2<br>
design: [2×58 double]<br>
channel: 'all'<br>
neighbours: []<br>
avgovertime: 'no'</font></p>
<p><font size="2">When I run ftStat = ft_freqstatistics(statcfg,
ftAllFiles{:}); , the program stops at the step 'findcluster':</font></p>
<p><font size="2">posclusobs = findcluster(tmp, connmat,
cfg.minnbchan);</font></p>
<p><font size="2">In fact tmp is 2D boolean table (29 frequency
samples x 401 time samples), connmat is 4x4 matrix with 0 and </font><font
size="2">cfg.minnbchan=0. So bug with </font><font size="2">findcluster.
See below. <br>
</font></p>
<p><font size="2">When I plot tmp (image(tmp)), the result doesn't
look ridiculous. <br>
<br>
</font></p>
<p><font size="2">How to have a good configuration of neighborhood
for this single-sensor time-frequency permutation test ? </font></p>
<p><font size="2">Thank you for your help. </font></p>
<p><font size="2">Best regards. </font></p>
<p><font size="2">Anne</font></p>
<p><font size="2">ps : <br>
</font></p>
<p>
<style type="text/css">.rtcContent { padding: 30px; }.lineNode {font-size: 10pt; font-family: Menlo, Monaco, Consolas, "Courier New", monospace; font-style: normal; font-weight: normal; }</style></p>
<p><font size="2">>> ftStat = ft_freqstatistics(statcfg,
ftAllFiles{:});<br>
the call to "ft_selectdata" took 0 seconds<br>
using "ft_statistics_montecarlo" for the statistical testing<br>
using "ft_statfun_depsamplesT" for the single-sample statistics<br>
constructing randomized design<br>
total number of measurements = 58<br>
total number of variables = 2<br>
number of independent variables = 1<br>
number of unit variables = 1<br>
number of within-cell variables = 0<br>
number of control variables = 0<br>
using a permutation resampling approach<br>
repeated measurement in variable 2 over 29 levels<br>
number of repeated measurements in each level is 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 <br>
the maximum number of unique permutations is 536870912<br>
generated 100000 random permutations<br>
computing a parametric threshold for clustering<br>
computing statistic<br>
estimated time per randomization is 0.00 seconds<br>
computing statistic 100000 from 100000<br>
<br>
Error using findcluster<br>
invalid dimension of spatdimneighbstructmat<br>
<br>
Error in clusterstat (line 214)<br>
posclusobs = findcluster(tmp, connmat, cfg.minnbchan);<br>
<br>
Error in ft_statistics_montecarlo (line 364)<br>
[stat, cfg] = clusterstat(cfg, statrand, statobs);<br>
<br>
Error in ft_freqstatistics (line 194)<br>
[stat, cfg] = statmethod(cfg, dat, design);<br>
</font></p>
<p><font size="2"> <br>
</font></p>
<p><font size="2"><br>
</font></p>
<pre class="moz-signature" cols="72">--
Anne Guérin-Dugué
PR Emérite Université Grenoble Alpes (UGA)
GIPSA-lab / PSD / Equipe ViBS
Bureau B146
Site Ampère
11 rue des Mathématiques
BP 46
F - 38042 GRENOBLE
tel : +33 (0)4 76 57 43 73
mel :anne.guerin@gipsa-lab.grenoble-inp.fr</pre>
</body>
</html>