Zanchenling Wang
Zanchenling Wang
Let me have a look
I just checked it out. We are good with the workflow. I just need to move numpy 2.0 to the build dependency list
I noticed the new bug. Let me figure it out.
Let me think about that. SciPy should be working as I remember that NNLS did pass the test on my own computer. I will double check.
Cheers! Fixed with #44 and #45, however I have noticed another issue that may prevent verbose mode from printing the right iteration step `i`.
One more thing in the todo list is update README for python 3.9~3.12
I'll take a look at the bug. I did the refactoring following the [`sklearn` NMF source code](https://github.com/scikit-learn/scikit-learn/blob/main/sklearn/decomposition/_nmf.py). They didn't do `fit().transform()` for `fit_transform`, so that additional `transform` step is avoided,...
I guess so. Though there is always something we can improve, I think it's good for now at least.
I have finished projection (both direct one and the L1-Normalization in Morup and Hansen) and PGDs for NumPy in wangzcl/archetypes@7b3d07d93279857108be9e2141c01579c015ffef. I will explain my design a little bit later.
`_projection.py`: I wrote an abstract base class `Projector` and two derived subclasses, `UnitSimplexProjector` (for the naive algorithm 1 mentioned in Condat, 2015) and `L1NormalizeProjector` (first non-negative, then normalize, mentioned in...