Feel free to close this if I've got this wrong (I'm a bit out of practice with Matlab but ran parts of this in octave), but I don't think the way you generate the loading matrix for the Lorenz datasets is working as intended.
Line 70 in Utils/generateDemoDatasets.m:
W = sort((rand(s, nCh, D)+1) .* sign(randn(s, nCh, D)), 1); % N x 3
would surely produce a weight matrix where all columns are sorted in ascending order, something like e.g.,
rather than a weight matrix that sorts according to the order of the first column only, e.g.,
which I'm guessing is how you would get 'neurons sorted according to their weighting for the first Lorenz dimension'
djoshea
commented
5 years ago
Feel free to close this if I've got this wrong (I'm a bit out of practice with Matlab but ran parts of this in octave), but I don't think the way you generate the loading matrix for the Lorenz datasets is working as intended.
Line 70 in Utils/generateDemoDatasets.m:
W = sort((rand(s, nCh, D)+1) .* sign(randn(s, nCh, D)), 1); % N x 3would surely produce a weight matrix where all columns are sorted in ascending order, something like e.g.,
rather than a weight matrix that sorts according to the order of the first column only, e.g.,
which I'm guessing is how you would get 'neurons sorted according to their weighting for the first Lorenz dimension'