Open axsk opened 6 years ago
I am happy to upgrade @axsk, I hold back only for the reason you mentioned, i.e. it takes some times for dependencies to catch up with backwards compatibility issues.
In principle, you should be able to use HMC, NUTS, MALA, SMMALA, since they can run both with reverse and forward mode autodiff. However, if there is such issue with ReverseDiff, we will have to wait I am afraid before I can upgrade Klara accordingly.
Actually I need to provide the gradients myself anyhow, so I don't really care for ReverseDiff. Too bad I cannot tell Julia to simply ignore the ReverseDiff issues..
Considering that ReverseDiff is not actively developed, maybe it should be removed from Klara anyhow?
ReverseDiff ist fixed now :)
Oh, great, thanks for letting me know. I will try to get Klara sorted in the next few days - I am away till Tuesday, so will look into this from Wednesday.
Any ETA of julia 0.7/1.0 releases?
Right now Klara is not usable under 1.0. One reason is the dependency on ReverseDiff, which is not yet ported as well :(
OT: Is there any gradient-based MCMC solution that I can use for now?