fix minimax's and negamax's best move search (works but might be inefficient)
some comments:
the MCTS and DFPN move-tests are commented out
There may be some bug in MCTS's evaluation function
updated DFPN's best_move selection (though still problematic)
pwyq@pwyq-X1C:~/github/Fill-Game$ ./fillgame_cli .*.*. 1
.
.
.
using DFPN...
W 0 0 1 0.000135 1
using minimax...
1
2 0 2
using minimax-ab...
1
2 0 2
using minimax-ab-tt...
1
2 0 2
using negamax...
1
2 0 2
using negamax-ab...
1
2 0 2
using negamax-ab-tt...
1
2 0 2
using pns...
1
1 0 1
using mcts...
terminate called after throwing an instance of 'std::invalid_argument'
what(): Cannot play at 1 0 with value 2, which is already filled. Full game = 2*2*1
Aborted (core dumped)
simulate()
game duplicationsome comments:
close #74