Hi, First at all, I would like to warmly thank the scikit developer community with providing us with such a high quality ML library: it really became an amazing piece of scientific software. I have a comment concerning the online documentation on Matrix Factorization Problems. (I use this mailing list because I could not find in your online howto, what is the best channel to communicate documentation issues.Apologies if this email is considered as spam in this mailing list !) On the webpage 2.5. Decomposing signals in components (matrix factorization problems) — scikit-learn 0.18.2 documentation We can read at 2.5.1.5. Sparse principal components analysis but a bit further, at 2.5.3.2. Generic dictionary learning, we can read The notations are obviously inconsistent as U and V have been interchanged some how. Two extra (less important) corrections could probably improve even further the clarity for the reader:1. Sticking to a single upper bound limit (either n_components or n_atoms)2. Specifying whether V_k are columns or rows (maybe using a notation à la Matlab/Numpy: V_{:,k} or V_{k,:}) Kind regards, Axel BREUER