[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Make your guesses

Posted By: Frank Berger
Date: Monday, 11 August 2025, at 5:48 p.m.

In Response To: Make your guesses (Stick)

Thanks for postings this interesting position ( XGID=----B-Ca----ABAa----A-C-AA:1:1:1:00:0:0:3:0:10 ).

I think two things we know for sure:

- if the player succeeds closing out the two checkers, he will win about 68%
- the player with the two checkers has nearly no decisions. Most moves are forced or there are very few alternatives.

John suggests trusting XG, but which numbers should be trusted?
Here are the numbers beginning with 2-ply ending with a rollout (1296 games, 3-ply move, XGR+ cube. Feel free to use any other setting).

Win Loose L-Gamm L-BG
2-ply 28,52 71,48 41,27 38,33
3-ply 16,86 83,14 68,03 55,75
4-ply 30,46 69,54 45,03 41,99
5-ply 20,42 79,58 58,44 45,92
XG-Roller 36,62 63,38 41,50 38,39
XG-Roller+ 21,06 78,94 57,63 43,55
XG-Roller++ 34,78 65,22 41,54 33,33
XG-Rollout 34,48 65,52 40,81 23,37

What you see is that the values vary strongly, e.g. although the win-percentage between XG++ and Rollout are stable, the BG differs by 10%!! So which value is the truth?

I check the evaluations of positions by evaluating on different levels. If the values vary much, IMHO the bot shouldn't be trusted too much.

A few will know that I claim that BGBlitz has an edge in containment games. So let's ask BGBlitz (I used the forthcoming TachiAI6, but with the current TAI5 it's probably not much different. If someone wants it right now, just ask).

Win Loose L-Gamm L-BG
2-ply 34,18 65,82 29,22 18,20
3-ply 42,20 57,80 26,10 14,35
4-ply 42,56 57,44 23,60 12,49
5-ply 46,34 53,66 22,49 11,81
BGB-Rollout 56,19 43,81 11,63 7,42

you see much less variation and the values move in one direction with higher ply . What I feel absolutely stunning is the result of the rollout. You might argue that if a bot has no idea of an position the results can't be trusted, but the side with two checkers has little possibilities to be played wrong, so how could misplay explain the results?

I was curious and played a half dozen of games (assisted by BGBlitz). In all games I were able to close out the two checkers so if we accept the 56% from above the closeout will succeed in about 80% which fits's much better than the 50% implied by XG 34% winning chance.

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.