Marc Lanctot

Great! Happy that the workaround worked Yeah it is due to a subtlety about how things get serialized. We should fix it, and we will, so please leave it open...

Hmm, I am not sure what's going on. I'll look into it. Tagging @elkhrt as I'm sure I will need his help on this.

Quick update that we are looking into it (and believed we have found the cause ... and by "we" I mean @elkhrt :)). We hope to have a fix out...

Hi @acxz, Thanks for the PR. There were compilation errors when trying to run the test. Can you take a look?

@acxz just checking in to see if you plan to continue working on this? If so, could I ask you to pull from master and push the merge commit to...

You can submit a PR, but merging them into the repos takes time and may never happen. (You can see a few examples in our current PRs.) For these cases,...

I am actually surprised to discover this because our MCTS implementations all handle chance nodes. @mlschmidt366 also implied that he had to do non-trivial changes to get games with chance...

Ok great, this should be easy to fix. Would you be willing to submit a PR if I walk you through what needs to be done? I would happy to...

Yep, whichever one you want to use. We can fix the other ones similarly later when we find the time. Thanks! I will write more in about 90 minutes.

Well the first one is the AlphaZero evaluator should be modified: if called on a chance node, then sample chance events until a normal node (or terminal) is reached, and...