Brendan O'Donoghue
Brendan O'Donoghue
Does the same thing happen for you when not running on GPU?
That's very strange because the main changes between those two tags are to the GPU solver (which I presume you're not using here) and adding the cmake install option (which...
To my eye those outputs look identical (except for timing etc). Could it be something changed inside cvxpy?
Thanks for posting this, can you try the steps [here](https://github.com/numpy/numpy/issues/15947#issuecomment-709516718) and see if that fixes it? It looks like downgrading numpy to 1.18 is the best approach now, either that...
Ok thanks for the update. I will see what I can do about this soon.
On the GPU stuff - the GPU code is still broken in some way that I haven't had time to track down. Overall the GPU doesn't provide much benefit anyway...
@syockit Thanks for this! I applied the patch and it worked! Do you want to turn this into a PR? The only problem I had was an erroneous 'infeasible' certificate...
Sure, no problem @syockit , thanks for sending in the patch!
This is probably just requires tuning `eps_abs` and `eps_rel` for more/less accuracy, and reducing `eps_infeas` to prevent erroneously returning infeasible, which appears to be what has happened. SCS has found...
Yes that's the idea, probably worth making sure that's actually better empirically though.