Sacha Chernyavskiy
Sacha Chernyavskiy
@lanctot the `torch` implementation tends not to converge (even though I basically tweaked basic hyperparameters: number of iterations, number of games played and the network size): the resulting exploitability is...
@lawrencewlcknight, did you make any attempts to fix that, or decided to leave the issue at it is?
there is also a need to tweak/upgrade the `jax` implementation as well, cause of haiku deprecation and strange `tf datasets` warnings, if what
It is not definitely hard not hard to do because it is mostly refactoring. However, to make the things work, we need a plan or roadmap just due to sheer...
hi, I forgot about the issue – go on
Hi @YashSachdeva. In spite of me fully sharing your intentions, I would still wait for any opinion of the maintainers first.
Yes, sure, I'll shortly try to be sure if it's possible to keep both, `pip` and `uv`, simultaneously supportable. Reason, why I asked in the fact that `uv` is being...
really, looks like a smth transient and easy to support: https://docs.astral.sh/uv/guides/migration/pip-to-project @hauntsaninja, does `uv` work nicely with external build systems, like `cmake`? or it is only for the python part?...
@lanctot wait with this one for one moment (till Monday), I want to check with the original paper, because it doesn't intuitively align why there shall be so many traversals...
hey @fuyuan-li, nice findings, the goal code is in fact quitre different from the paper. I decided that it might be a good thing to work on both, `jax` and...