Error with transpose/permute

Hello everyone!! I have this doubt about the use of permute. I have here a simulated hemodynamic response (simulating an task evoked fMRI) and I applied to these response a certain delay/lag. After this, I want to apply glmfit function to this hemodynamic response with the applied lag (my X) and to a 4D matrix (my Y), with this matrix simulating data from an fMRI. I didnt put the fMRI data here because you needed to download, so I used this matrix to simulate that data and to try to figure it out what´s going on. Basically the error is the following ""Error using ' Transpose on ND array is not defined. Use PERMUTE instead.", and I'm not even using the transpose function...

Please help me!!

Thank you so much for your time!

% import fMRI data (you can find this file on the internet)
 data = load('P05120.7-MC.mat');

%expected hemodynamic response
hemo=[repmat(0,10,1);repmat(1,10,1);repmat(0,10,1);repmat(1,10,1);repmat(0,10,1);repmat(1,10,1);repmat(0,10,1);repmat(1,10,1);repmat(0,10,1);];

% find our delay (lag) through cross-covariance 
voxel = data.Vfunc(27,37,9,:);
new_voxel = reshape(voxel, [], 1);
[c lag] = xcov(new_voxel, hemo, 'coeff'); 

%applying lag to our hemodynamic response
delay = 0;
for z=1:length(c)
    [max_cov, idx] = max(c);
        if idx == z
            delay = lag(z);
        end
        if delay > 0
            hemo_lag = [(padarray(hemo(1:length(delay)-1),[delay,0],0, 'pre')); hemo(1: length(hemo)-delay)]; %aplicar lag à resposta hemodinamica
        end
end

% applying glmfit function to every elements of the matrix 
 matriz = randn(2, 3,3,2);
 m = permute(matriz,[2,1,3,4]);

for i=1:3
    for j=1:2
        for k=1:3

            [params, dev, stats] = glmfit(hemo_lag(:), m(i,j,k,:), 'normal');
            
        end
    end
end

[Rik]

What is the full error message? It is unclear which command in your code triggers the error.

[Iugo]

Hi Rik! The full error message is this: Error using '

Transpose on ND array is not defined. Use PERMUTE instead.

Error in statslib.internal.removenan (line 38)

y = y';

Error in internal.stats.removenan (line 30)

[badin,wasnan,varargout{1:varargoutsize}] = statslib.internal.removenan(varargin{:});

Error in statremovenan (line 7)

[badin,wasnan,varargout{1:nargout-2}] = internal.stats.removenan(varargin{:});

Error in glmfit (line 190)

[anybad,wasnan,y,x,offset,pwts,N] = statremovenan(y,x,offset,pwts,N);

Error in project18_4 (line 36)

[params, dev, stats] = glmfit(hemo_lag(:), m(i,j,k,:), 'normal');

Answer 1

[Walter Roberson]

https://www.mathworks.com/help/stats/glmfit.html

In most cases, y is an n-by-1 vector of observed responses. For the binomial distribution, y can be a binary vector indicating success or failure at each observation, or a two column matrix with the first column indicating the number of successes for each observation and the second column indicating the number of trials for each observation.

 matriz = randn(2, 3,3,2);
 m = permute(matriz,[2,1,3,4]);

Those are 4 dimensional.

[params, dev, stats] = glmfit(hemo_lag(:), m(i,j,k,:), 'normal');

i, j, k are scalars, so m(i,j,k,:) is 1 x 1 x 1 x 2 . Something that is 1 x 1 x 1 x 2 is not an n x 1 vector or a array with two columns.

I suggest

squeeze(m(i,j,k,:))

[Iugo]

Thank you for your help and time Walter!

So I just eliminate this portion below?

 m = permute(matriz,[2,1,3,4]);

I just did what you said and reported this error:

Error using glmfit (line 194)
Number of observations in X and Y must match.

Error in project18_4 (line 39)
            [params, dev, stats] = glmfit(hemo_lag(:), squeeze(m(i,j,k,:)), 'normal');

[Walter Roberson]

No, if you only did that, then your line

[params, dev, stats] = glmfit(hemo_lag(:), m(i,j,k,:), 'normal');

would become

[params, dev, stats] = glmfit(hemo_lag(:), m(j,i,k,:), 'normal');

but you would still have the problem that that would be a 1 x 1 x 1 x 2 vector.

You need

[params, dev, stats] = glmfit(hemo_lag(:), squeeze(m(i,j,k,:)), 'normal');

[Iugo]

I did what you said:

matriz = randn(2, 3,3,2);
m = permute(matriz,[2,1,3,4]);

for i=1:3
    for j=1:2
        for k=1:3

            [params, dev, stats] = glmfit(hemo_lag(:), squeeze(m(i,j,k,:)), 'normal');
            
        end
    end
end

And still gives me the error:

Error using glmfit (line 194)
Number of observations in X and Y must match.

Error in project18_4 (line 39)
            [params, dev, stats] = glmfit(hemo_lag(:), squeeze(m(i,j,k,:)), 'normal');

Sorry for bother you again...

[Walter Roberson]

Your code does not define hemo_lag . You assign to hemo_lag if delay > 0 but you assign 0 to delay so that assignment is never done. We have to presume that hemo_lag was loaded in with the file. The assignment you do hints strongly that hemo_lag is a lot larger than size(m,4), which is 2.

Your assignment to matriz should very likely have at least one dimension whose size depends upon the number of elements in hemo_lag -- unless you can be sure that the data will only be a particular size (and if that is required, then toss in an assert() on the size, just-in-case.

[Iugo]

So what do you suggest? I'm not quite understanding what to do...