You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Apologies for three issues, but I figured I should post minimum reproducible examples for three issues I've run into.
Suppose I want to do sparse non-negative matrix factorization on the matrix $(1, 2) \ (2, 4)$. I run: fitted(nmf(matrix(c(1, 2, 2, 4), nrow = 2), 1, 'snmf/r'))
I would like to get the perfect rank-1 approximation $(1 \ 2) * (1, 2) = (1, 2) \ (2, 4)$. However, SNMF returns $(0.92, 1.84) \ (1.84, 3.68)$.
I assume there is something about the SNMF $L_1$ penalty that pulls the predictions towards zero. Is it clear what argument to change, and how, to get better fit under SNMF?
The text was updated successfully, but these errors were encountered:
Apologies for three issues, but I figured I should post minimum reproducible examples for three issues I've run into.
Suppose I want to do sparse non-negative matrix factorization on the matrix$(1, 2) \ (2, 4)$ . I run:
$(1 \ 2) * (1, 2) = (1, 2) \ (2, 4)$ . However, SNMF returns $(0.92, 1.84) \ (1.84, 3.68)$ .
fitted(nmf(matrix(c(1, 2, 2, 4), nrow = 2), 1, 'snmf/r'))
I would like to get the perfect rank-1 approximation
I assume there is something about the SNMF$L_1$ penalty that pulls the predictions towards zero. Is it clear what argument to change, and how, to get better fit under SNMF?
The text was updated successfully, but these errors were encountered: