Alexey Stukalov
Alexey Stukalov
> I made the changes you suggested with a few adjustments. Thank you! > The method parameter can now take :classic, ':normalized, and :adjusted`. I think it is better to...
Another random failure in COO matrix-matrix multiplication: https://github.com/JuliaSparse/MKLSparse.jl/actions/runs/10893269544/job/30227846794?pr=52 This time it is in `mkl_sparse_z_mm()` (last time was `mkl_sparse_d_mm_64`)
Maybe the core method for the distance API should be `dist(T, x, y)` with `dist(x, y) = dist(result_type(x, y), x, y)`
That will help with `colwise!()`-like methods, where the eltype of the result is known in advance (but currently it is ignored): `colwise!()` will call `dist(T, x, y)` directly. In other...
These matrices might be large, so promoting Int16 to Float64 just to tell `colwise!()` how to calculate the distance (which it is currently doing element-wise) is a bit of an...
Thank you for the detailed explanations! Implementing the workaround in the user script is probably the optimal strategy now, but look at it from the perspective of the other packages,...
@pfitzseb @davidanthoff Could you please review this PR?
@pfitzseb @davidanthoff Any chance this PR could be reviewed?
Thank you for the report! Can you submit the PR with the fix and the tests case (D >= N) confirming that the algorithm works correctly?
@AzamatB Thank you for the fix and sorry for very long delay in reviewing it! Do you have a test case for the fix?