Rémi Louf
Rémi Louf
I pushed the change in #344 and cut a new release; we'll just have to check the tarball once 0.9.1 is published on Pypi.
That's awesome, thank you! Can I and @junpenglao be made admin of the repo?
I opened two PRs, you should have gotten notifications. Otherwise: https://github.com/conda-forge/blackjax-feedstock/pulls Thank you again!
Want to do it too?
Not that you need it, but it should be trivial based on this: https://github.com/blackjax-devs/blackjax/discussions/360#discussioncomment-3756412
Please do! It would be great to add the binary neural network as well.
A few thoughts: - I'm on the fence when it comes to keeping CV and implementing SVRG; not only do they require to keep track of a `GradientState`, they also...
> High level question: here you also revert the change in https://github.com/blackjax-devs/blackjax/pull/293/ where you turn step size into a callable (so user can control how step size change during sampling),...
@junpenglao @bstaber The behavior in a loop was not very clear indeed, so let me give a full example. I modified (hopefully improved) the behavior slightly. For the simple Robbins-Monro...
@junpenglao This is ready for review. I implemented the control variates as a wrapper around the simple estimator, as dicussed above. The code is much cleaner than before, and svrg...