add chess implementation

[yeekit24]

Resurrect PR Chess prototype #147 by @namin

• now it did not stop at self.play stage, as i re-implement the function getGameEnded

def getGameEnded(self, board, player):
        # return 0 if not ended, 1 if player 1 won, -1 if player 1 lost
        outcome = board.outcome()
        if outcome is not None:
            if outcome.winner is None:
                # draw return very little value
                return 1e-4
            else:
                return who(outcome.winner)
        return 0

• fixed a bug in Arena.py where it showed some weird behaviour.
• added stockfish player.

Once again, thanks @namin for the code implementation.

[namin]

Glad you're doing this! Have you managed to train a decent chess bot with this?

[dweekly]

This seems to precisely invert the logic of which party won; was this done to fix a bug?

[jamesbraza]

I think the original code is correct, this was not a bug.

Note a few lines up the order of the players was inverted, and also in Arena.py it normalizes for player ID via curPlayer * self.game.getGameEnded(board, curPlayer) here.

[dcy11011]

This may be because the getGameEnded(self, board, player) method of ChessGame returns 1 if player 1 won instead of the given player won. Suggest multiplying the return value in ChessGame.getGameEnded by the function parameter player, instead of modifying the code here.

[jamesbraza]

I think this would break Coach.py for all other games

Note the same comment holds for the change in MCTS.py

[jamesbraza]

I think the original code is correct, this was not a bug.

Note a few lines up the order of the players was inverted, and also in Arena.py it normalizes for player ID via curPlayer * self.game.getGameEnded(board, curPlayer) here.

[jamesbraza]

I think we can from utils import AverageMeter, where is pytorch_classification.utils.Bar though?