almostimplemented / checkers Goto Github PK

Maybe a minor bug. I noticed in the game I had 2 jump choices. One was the start of a double jump. I did the first jump of the double jump. Instead of forcing me to complete the double-jump, it allowed me 2 jump choices: 1) completing the double jump, or 2) taking the other jump choice I had at the beginning.

I'm far from a rules expert, but the one set I saw said you have to complete multiple jump sequences.

I attached dump of the game. Starts at turn 50.

my_game.txt

When I run game.py with either arthur or random_agent, I get the following error :

Traceback (most recent call last):
  File "game.py", line 237, in <module>
    status = main()
  File "game.py", line 168, in main
    B.make_move(cpu_1.make_move(B))
  File "C:\Users\julro\Desktop\Coding\DeepCheckers\checkers-master\checkers.py", line 66, in make_move
    taken_piece = int(1 << sum(i for (i, b) in enumerate(bin(move)[::-1]) if b == '1')/2)
TypeError: unsupported operand type(s) for <<: 'int' and 'float'

It doesn't always occur on the same line but it's always that same TypeError

Hey could you please update license, so I can improve this codebase?

Thanks for posting this code. I'm thinking of using this as checkers environment for an implementation of AlphaZero agent I'm working on.

1. As far as you know, does this environment correctly implement all the rules?

2. Regarding the agent, I've always been terrible at checkers and haven't played since I was a kid. I beat the agent on my first try. Should it be relatively strong? Perhaps I got lucky.

Thanks a lot, Pat

Arthur Samuel's program was historically the first learning agent.

Here, I tried to play a game against the arthur agent, with the python2 game.py command. Followed by selecting arthur as the agent. I typed 0 all the way through. Always selecting the first move. And I won !!

I'm under the opinion that the arthur agent hasn't been learning much... it is supposed to be on par or defeat average checkers players. And, here, 0 all the way through, and I win.

I also do not see many things in the code that could be construed as implementing the kind of prototype reinforcement learning that Arthur Samuel pioneered. Nor in the documentation.

So. Question: is learning implemented ?