RFC Make non_negative_factorization call NMF instead of the opposite #19607
Conversation
|
Hello everyone I need some help about from the lines I put here a snippet having the same problem import numpy as np
from sklearn.decomposition import non_negative_factorization
rng = np.random.RandomState(0)
X = rng.random_sample((20, 15))
H = rng.random_sample((15, 15))
W = rng.random_sample((20, 15))
non_negative_factorization(X, H=H, update_H=False)Apparently the scikit-learn/sklearn/utils/validation.py Line 1040 in 4beb0c2 (I have temporarily added a print(attrs) to check). Is that an expected behavior? Thanks!
|
Let's forget the idea of being able to call transform without fit. It makes the code too complicated. That way this PR is completely transparent to the user and doesn't change the API at all. I directly pushed this. |
|
Thanks @cmarmo ! |
Reference Issues/PRs
Working on the MiniBatch implementation of the Non-negative Matrix Factorization class (#16948), it has been suggested that a reformat of both standard NMF and the in progress MiniBatch implementation would improve the maintainability of the code. This pull request applies this reformatting using #14985 and #18975 as models.
What does this implement/fix? Explain your changes.
Make
non_negative_factorizationcallNMFinstead of the opposite.Any other comments?
I had to add a tag to the
NMFclass to avoid then_features_incommon test: but I'm not sure this was the right way to make the test pass.ping @TomDLT and @jeremiedbb : thanks!