Closed ghbrown closed 1 week ago
@ghbrown Can you please add an issue as described in the GitHub workflow in CONTRIBUTING.md? And please include a motivating use case for this so that is captured in the Issue.
Be careful about the convergence check which uses the last MTTKRP (at least in MATLAB) and assumes it's a particular mode that is left out.
Thanks for the warning, Tammy. I just looked over that part of the code, and because of where and how I updated dimorder
I believe all of that should be properly taken care of automatically.
But I did catch and fix a comment that needed to change!
An argument called
optdims
of typeList[int]
is added tocp_als()
to enable optimizing only the factor matrices of specific modes. This was not previously possible even by scripting aroundcp_als()
since it checks for convergence only after each full sweep. Control flow was hardly altered (dimorder
may now just be missing some entries), and the default is of course optimizing all modes.📚 Documentation preview 📚: https://pyttb--302.org.readthedocs.build/en/302/