Twixt first move philosophy TWIXT PP
5 replies. Last post: 2019-04-09
Reply to this topic Return to forum-
Alan Hensel at 2019-04-08
A long, long time ago, I can still remember, I think it was 14 years ago to be precise, I wondered if the standard first moves were really the best.
Back then, the dominant Twixt culture was to start with “1.d3” or “1.e3”. But the were tens of thousands of games; surely some statistical analysis could provide a more solid idea of what the best first move is? Well, I ran those stats, and I hope that got people's first moves a little bit unstuck and gave people the courage to try some different first moves. Those stats have recently moved here.
“1.d3” and “1.e3” seem to be considered way too weak by the bot. The bot seems to think row 6 (and its symmetrical reflection 19) is the most even-handed row, and horizontal positioning doesn't seem to matter much. I have no insider info on the bot; I just looked at the existing games played by the bot, again by hand, no automation (shame on me again!) and recorded them here.
-
TwixtBot at 2019-04-08
TwixtBot's first/second move strategy is different than all other moves. I didn't want to figure out how to make the AI deal with the swap rule, so I built a simple strategy. First, I played the bot against itself a few dozen times from each possible first move with no swap. I now have noisy approximations of how often white wins “unswapped” from each point. I applied some smoothing to the statistics using the theory that the closer to the center that first move is, the more likely it is to be a winner.
The second move is now easy: if the win rate according to my smoothed data is >50% for white, I swap. For the first move, rather than pick the one point closest to 50%, I decided to make a weighted random selection: the closer the win rate is to 50%, the higher the chance I pick that move. As you noted, the surprising thing is that the column barely matters; it's almost entirely the choice of row that counts.
We can sort of see that the AI agrees with this assessment by noting how often it plays freakishly close to the edge in the early game.
-
David J Bush ★ at 2019-04-09
I'm not sure I agree with the premise that a first move closer to the center but on the same row would be stronger. L5 seems considerably weaker than E5, for example. E5 can be a ladder catcher, both in attack and defense, but what good is L5? It's difficult to argue with results, though.