Brendan O'Donoghue
Brendan O'Donoghue
Just uploaded version 1.2.7.
Sorry, I can't read those error messages so I can't tell what's happening. In many cases the GPU solver is actually slower than the CPU solver, so I would advise...
Looks like you got it to work? By the way, the SCS python CI is with the interface which I broke out to a separate repo, it's here now: https://github.com/bodono/scs-python
Yes, by default it links against the numpy blas / lapack. I have generally found conda to be the easiest way to get it all running. You should just be...
Thanks for reporting such a detailed problem. The issue here is that SCS is a first-order solver and as such is good at getting moderate accuracy quickly, but possibly taking...
SCS will terminate if it either hits `max_iters` or the tolerance is below `eps`, so you might need to increase `max_iters` and decrease `eps` to get the accuracy you want....
Also setting `verbose=True` will allow you to see why SCS has decided to terminate and eye-ball the progress toward the end.
Thanks for the question Thomas. You're right this isn't documented anywhere, and I will leave this issue open so I can remember to document carefully all the parameters. The rho_x...
I would keep `rho_x` small and actually play around with the `scale` parameter, which makes much more difference in practice. The rule of thumb is that if the primal residual...
Thanks for posting this. Has this improved the performance sufficiently? If not you should try modifying the scale parameter as I suggested. I realize you are not using the built-in...