gotta fix the iterative deepening algorithm for the endgame. it's trying to stick you and looking deeper and deeper (generating a ton of 1-tile plays) without realizing that RATLINS wins by more. There's no super easy way to do it but one thing I can do is something like "if it's found the same play 4 plies in a row just play it" or whatever.
@@cesardelsolar3971 I have no idea how the bot works, but I assume it is something like, sort all options by score + static, then simulate full games a la quackle for the top X options and finally play the one with the highest win%?
@@Goryus for most of the game yes. When there is one or fewer tiles in the bag it switches to solve the game exhaustively. The exhaustive solver is trying to work out every possible combination to deeper and deeper depths to see if it does better than RATLINS. I need to give it a shortcut.
@@cesardelsolar3971 if it's the case that it is simulating full games after identifying candidate moves, something that might be worth trying (if you haven't already) is simulating just the next 2 ply instead. Rather than win rates, you can then use the score averages for those 2 ply (subtracting out the opponent's average score, of course). I have a suspicion (based on some experience with Quackle's sim option) that simulating full games may have the effect of "washing out" the impact, making very dangerous plays appear safer than they really are because the damage gets averaged out after many moves. There is also bias added by the simulation that in principle will add up the deeper you go. I suspect that a shallow but very wide search may be stronger in this specific case than a deep but narrow one. Please pardon the unsolicited commentary from an uninvolved bystander, and I've loved watching the games with your bot.
yeah, if we assume that the bot is likely to have something like OX on its rack, then OFFSIdE just lets it play O14 OX followed by a 15A play and possibly a 7 next to the O after all the bot has been exceptionally strong at creating bingo racks for itself
spoiler blocker
If we just all call the next game 22 again and forget about the first 10, Mack would be up 12-9
Lol... if only!
gotta fix the iterative deepening algorithm for the endgame. it's trying to stick you and looking deeper and deeper (generating a ton of 1-tile plays) without realizing that RATLINS wins by more. There's no super easy way to do it but one thing I can do is something like "if it's found the same play 4 plies in a row just play it" or whatever.
Thanks for the explanation! All makes sense, except the part about "RATLINS wins by more" -- hey I still won the game 😛
@@mackmeller aha, the endgame solver just tries to maximize spread, I think that’s what I meant :) congrats on the recent comeback!
@@cesardelsolar3971 I have no idea how the bot works, but I assume it is something like, sort all options by score + static, then simulate full games a la quackle for the top X options and finally play the one with the highest win%?
@@Goryus for most of the game yes. When there is one or fewer tiles in the bag it switches to solve the game exhaustively. The exhaustive solver is trying to work out every possible combination to deeper and deeper depths to see if it does better than RATLINS. I need to give it a shortcut.
@@cesardelsolar3971 if it's the case that it is simulating full games after identifying candidate moves, something that might be worth trying (if you haven't already) is simulating just the next 2 ply instead. Rather than win rates, you can then use the score averages for those 2 ply (subtracting out the opponent's average score, of course).
I have a suspicion (based on some experience with Quackle's sim option) that simulating full games may have the effect of "washing out" the impact, making very dangerous plays appear safer than they really are because the damage gets averaged out after many moves. There is also bias added by the simulation that in principle will add up the deeper you go. I suspect that a shallow but very wide search may be stronger in this specific case than a deep but narrow one.
Please pardon the unsolicited commentary from an uninvolved bystander, and I've loved watching the games with your bot.
OFFICE for 45 just blows everything else out of the water in my opinion
yeah i am expecting something on the scale of ONYX
yeah, if we assume that the bot is likely to have something like OX on its rack, then OFFSIdE just lets it play O14 OX followed by a 15A play and possibly a 7 next to the O
after all the bot has been exceptionally strong at creating bingo racks for itself
well done yes you can win it for us fleshies before our AI overlords take over.
Hopefully!