Closed jolars closed 2 years ago
the proxsplit method can be improved and it's really good to look into it so no worries. It's really great to have access to the time like this, it was missing. Thanks a lot!
I rename the variable times
to avoid clash with the time module, and also because it stores more than one times (the same way we he gaps)
Since we return one more variable now, I deleted theta from the returned variables of prox_grad
LGTM !
Great, thanks @mathurinm !
And thanks @Klopfe too! (Sorry I missed your reply at first)
Add some lines to collect timings from inside solvers. I considered making this optional but I don't really think the overhead is large enough to care about, or what do you think?
Here's the current state of affairs (pgd is actually fista). And yes, you don't need to tell me about how "well" the proxsplit strategy performs =,)