Open simondorfman opened 3 years ago
WIP
Extremely interesting.
As I'm playing the computer, it frequently happens that only a small handful of moves don't instantly lose (with best play).
This is super-interesting. I'm learning a lot, playing this way.
Couple things I noticed:
I don't suppose you have a way to reproduce that? I saw it once, but haven't figured out how to reliably do it again.
Actually I think I might have found it... For some reason I had a max iteration check in addition to the max time, so the analysis engine was terminating it's search and continuing the game as if it made the previous move it was evaluating.
Also, I'm adding a blue color to highlight the best move.
OK, I ended up fixing 5 different bugs for the overlay (which together should fix all 4 issues you mentioned above).
@skotz This is interesting. Blue is about to win after taking this turn, regardless of play, just from growth that's about to happen. This seems to confuse the AI. It thinks the two moves that would win before growth are the worst moves and everything else is the best move. In this case, it doesn't really matter, but maybe this illustrates some flaw in the logic somewhere...
Game transcript:
N21 N09 G S32 N02 G S14 S28 G S30 N23 G S34 N08+ G S30+ N23 G N32 S33+ G N24+ S34 G S34 N24 G N24+ S21 G
Odd
use an engine for every possible move and give it a value maybe even a heatmap overlay, showing the tiles x configuration away from that tile
This will help humans work backwards about which moves are stronger, then we can feed it back to the engine
Basically use the Evaluation number from the Engine Output.
Have the engine run indefinitely and keep updating the numbers on the overlay.