I think it may be that the calculation is falling back to matrix multiplication, which does not raise an error because our domain is square (nrows==ncols), but this should be element-wise multiplication, right? I think it is at least safe to explicitly name it as element-wise:
Getting some weird results sometimes, with the channel width computation here.
I think it may be that the calculation is falling back to matrix multiplication, which does not raise an error because our domain is square (nrows==ncols), but this should be element-wise multiplication, right? I think it is at least safe to explicitly name it as element-wise:
Agree?