NMF returns all-zero weights when scikit-learn updates to 0.24.2 (or newer)
Question
I used Non-Negative Matrix Factorization (NMF) to conduct topic modelling for a long time.
When I updated my environment to the latest sickit-learn update (from 0.24.1 to 0.24.2) NMF stopped working properly and instead returned all-zero weights.
It is worth noticing that this error did not apply for the NMF-Kullback variation.
My NMF training script looked like this:
nmf = NMF(n_components = self.no_topics, alpha = 0.1, l1_ratio =.5, init = 'nndsvd').fit(tfidf)
Any ideas?
1 answers
solution1
0 2021-12-13 12:29:04
2 ways of resolving this problem worked for me:
- Setting alpha equal to 0.0 (sub-optimal)
- Downgrading sickit-learn back to 0.24.1
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.
Related Question
custom matrix in NMF in scikit-learn
Scikit-learn updates in Python
Using scikit-learn NMF with a precomputed set of basis vectors (Python)
Scikit-Learn Classification and Regression with Weights
Scikit-learn: role of weights in Ridge Regression
Scikit-learn pipeline returns list of zeroes
sample weights in scikit-learn broken in cross validation
Scikit-learn non-negative matrix factorization (NMF) for sparse matrix
Nearest Neighbor using customized weights on Python scikit-learn
what's the use of transformer_weights in scikit-learn pipeline?
Related Tags