Michael Pang
Michael Pang
Yeah my impression is anything we understand won't matter anyway :) All we can do is ensure the inputs and outputs are correct and pray for the best. BTW are...
Try checking the [new branch](https://github.com/Akababa/chess-alpha-zero/blob/opts/src/chess_zero/agent/player_chess.py), I removed that part and optimized a lot of other stuff.
@benediamond sorry I didn't see your other comment. I think Python passes the reference to the [] to self.prediction queue so it's all good. You can uncomment the ` #logger.debug(f"predicting...
By the way, I'm brainstorming a [list of ways](https://github.com/Akababa/chess-alpha-zero/wiki/Preventing-draws-by-repetition) to fix the draws by repetition thing, hopefully we can figure this one out.
Thanks, if that works it's a much nicer solution than the stuff I came up with. Did you let tau=e^{-0.01*turn} ?
What version? I'm on cuda 8 and cudnn 6
I just used the old versions on the [TF site](https://www.tensorflow.org/install/). Are the new ones faster?
I hear you; I was [trying to do that](https://stackoverflow.com/questions/47929533/how-to-allocate-pooled-resource-with-processpoolexecutor-in-python) for the longest time before giving up and using shared manager. The processpool will only have 3 running at a time,...
Yeah python is slow but since I already maxed out my GPU usage I will just keep using it for now. When we do distributed self play we will most...
Nice, I look forward to seeing it! I'm not sure what your question is. The SL policy is just 1 for the human move and 0 for everything else. And...