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
We need an init_fn for nonnegative factorization (this can simply be a wrapper to nnmf) and for nonnegative loadings (with arbitrary factors). (And similarly, for nonnegative factors.) For the latter, I have been running the default udv_si and then using either u and v or -u and -v depending on whether the negative or positive part of u is larger (in a L_2 sense, which seems to work better than L_1, but maybe L_\infty could also work well). I don't whether this is the best way to do it.
The text was updated successfully, but these errors were encountered:
We need an
init_fn
for nonnegative factorization (this can simply be a wrapper tonnmf
) and for nonnegative loadings (with arbitrary factors). (And similarly, for nonnegative factors.) For the latter, I have been running the defaultudv_si
and then using eitheru
andv
or-u
and-v
depending on whether the negative or positive part ofu
is larger (in a L_2 sense, which seems to work better than L_1, but maybe L_\infty could also work well). I don't whether this is the best way to do it.The text was updated successfully, but these errors were encountered: