Inefficient processing of cones about cvxr HOT 7 CLOSED

@aszekMosek I've made some progress. Of course, I need to exhaustively check my patches carefully and roll up a release, but I attach the results here, just FYI. (BTW, CVXR hasn't caught up to the newer CVXPY releases, but that is in progress...) This also fixes all the warnings about inefficient matrix coercions...

Some timing result comparisons below: old, followed by new.

   user  system elapsed 
 35.150   7.613  42.793 

  user  system elapsed 
  0.873   0.062   0.990 

cvxr_issue137.txt

from cvxr.

Great, thank you both. I will close this ticket.

from cvxr.

@aszekMosek Thank you for the report. Indeed, we have seen some issues like this and Anqi and I are working on a new release. But I will investigate this further as soon as I can.

from cvxr.

@bnaras that sounds great!

Would you be willing to post/share your current patch? Or is the next release coming so soon that it's worth waiting?

I have been experiencing similar performance issues with CVXR. I'd prefer to work with CVXR. But I'd switch to CVXPY if the fix to CVXR needs some time (say weeks rather than days).

from cvxr.

@florianpeters I've merged the changes into the master branch (version 1.0-13 of CVXR). CRAN upload is in process.

from cvxr.

Version 1.1-13 is now on CRAN and the builds for platforms will soon appear.

from cvxr.

@bnaras fantastic! The new release works amazingly well on @aszekMosek's example and significantly improves performance on more complex optimization problems as well. Thanks for rolling out the new release so quickly!

from cvxr.