Assignment error in clusterSingleBatches for very few channel data

[Jialiang-Lu]

Hi,

I need to spike sort my single channel recordings. Following the advice on the website, I am running Kilosort2 on four channels (i.e. only four channels have connected==true) where three of them are just noise. Unfortunately, preprocessDataSub believes that two channels are bad and my ops.Nchan becomes 1, which lead to Nfilt==1 in clusterSingleBatches and the error:

Error using gpuArray/subsasgn
Assignment has more non-singleton rhs dimensions than non-singleton subscripts.
Error in clusterSingleBatches (line 93)
        Whs(:, ibatch)     = int32(Wheights);

The error happens because Wheights is 1x2 array and ibatch is a scalar. Wheights is caculated here:

(72)      W = dWU./(1e-5 + mu);
(73)      
(74)      W = reshape(W, nPCs, Nchan, Nfilt);
(75)      nW = sq(sum(W(1, :, :).^2,1));
(76)      W = reshape(W, Nchan * nPCs, Nfilt);
(77)       
(78)      [~, Wheights] = max(nW,[], 1);

My questions is, what does the sum on the first dimension do on line 75 if W(1, :, :) has size 1 along the first dimension? If W is nPCs x Nchan x Nfilt on line 74, then nW would be an Nchan x Nfilt array (if Nfilt>1) or 1 x Nchan array (if Nfilt==1 in my case). Wheights would have the same dimension as nW and would never have the second dimension==1, therefore cannot be assigned to Whs(:, ibatch).

Am I missing something important, or is it the case that only two good channels simply wouldn't work?

[Jialiang-Lu]

Ok, making ops.minfr_goodchannels = 0 solves the problem, at least for now.