BGonline.org Forums

How To Trap An Anchor Bot Comparison Part 1

Posted By: Michael Depreli
Date: Sunday, 2 May 2010, at 2:56 p.m.

This study is a continuation of Tim's thread How To Trap An Anchor
I wanted to get a feel how the bots (GNU & XG) played this position at various plies and hopefully learn something in the process.
This is part one and looks at the bots' responses assuming blue rolled 21 9/6 (a reasonable bot choice) and white responded with a roll that didn't change the position.
Part two will assume blue played 8/6 8/7 and white rolled 32 which I will post in a few days.
In the table are best moves as selected by GNU 2-Ply, 3-Ply and a 2-Ply Rollout plus XG 3-Ply Huge, 3-Ply Forced, 4-Ply Huge and XG 3-Ply Rollout.
Error rates are taken based on both GNU and XG rollouts and totalled.
Rollout parameters were:
GNU 2-Ply 1296 trials 24 moves within 0.20 stop at JSD 4.0.
XG 3-Ply Huge 1296 trials.
ALL ROLLOUTS AND ERROR RATES ARE CUBELESS
At this stage I only wanted to get a general feel rather than absolute accuracy hence the low number of trials. If there's sufficient interest, I'll happily do stronger rollouts.
The first thing that struck me was how many of XG's 3-Ply choices changed when forcing it to evaluate all moves past the huge filter. Note also that the forced evals (red font) had a ZERO error rate based on XG and GNU rollouts. This could be problematic in that if we use XG 3-Ply Huge for rollouts it clearly won't be playing optimally. One reason why I have asked Xavier for configurable search filters.
So we could use 4-Ply huge for rollouts instead but bizarrely 4-Ply huge based on this small sample has a poor error rate.

For this small sample it appears 3-Ply Forced would be slightly ahead of GNU 2-Ply & 3-Ply. Shame we can't utilise it.

I'd be very interested in hearing expert commentary on whether you agree with the bots' choices.

	GNU	Error	Error	GNU	Error	Error	GNU	XG3 RO	XG	Error	Error	XG	Error	Error	XG4	Error	Error
Roll	2-Ply	G R/O	XG R/O	3-Ply	G R/O	XG R/O	2-Ply R/O	3-Ply R/O	3-Ply Huge	G R/O	XG R/O	3-Ply Forced	G R/O	XG R/O	4-Ply	G R/O	XG R/O
66	8/2(2) 7/1(2)	0.000	0.000	9/3(2) 8/2(2)	0.039	0.010	8/2(2) 7/1(2)	8/2(2) 7/1(2)	10/4(2) 9/3(2)	0.065	0.050	8/2(2) 7/1(2)	0.000	0.000	8/2(2) 7/1(2)	0.000	0.000
65	8/3 8/2	0.026	0.004	9/4 9/3	0.043	0.033	16/10 6/1	16/10 7/2	19/8	0.026	0.023	19/8	0.026	0.023	9/4 9/3	0.043	0.033
64	16/10 7/3	0.010	0.021	16/10 8/4	0.038	0.049	8/4 8/2	8/4 8/2	16/6	0.013	0.009	16/6	0.013	0.009	8/4 8/2	0.000	0.000
63	16/10 6/3	0.000	0.000	16/10 6/3	0.000	0.000	16/10 7/3	16/10 6/3	16/7	0.022	0.038	16/7	0.022	0.038	16/10 7/4	0.005	0.014
62	16/8	0.016	0.024	16/10 6/4	0.000	0.000	16/10 6/4	16/10 6/4	16/8	0.016	0.024	16/10 6/4	0.000	0.000	10/8 10/4	0.124	0.115
61	16/9	0.000	0.000	16/9	0.000	0.000	16/9	16/9	16/9	0.000	0.000	16/9	0.000	0.000	8/7 8/2	0.042	0.037
55	8/3(2) 7/2(2)	0.000	0.000	9/4(2) 8/3(2)	0.031	0.022	8/3(2) 7/2(2)	8/3(2) 7/2(2)	9/4(2) 8/3(2)	0.031	0.022	8/3(2) 7/2(2)	0.000	0.000	9/4(2) 8/3(2)	0.031	0.022
54	8/4 8/3	0.009	0.000	8/4 8/3	0.009	0.000	7/3 7/2	8/4 8/3	16/7	0.009	0.011	16/7	0.009	0.011	8/4 8/3	0.009	0.011
53	16/11 6/3	0.010	0.024	16/8	0.005	0.032	7/4 7/2	7/4 7/2	16/8	0.005	0.032	16/8	0.005	0.032	7/4 7/2	0.000	0.000
52	16/9	0.011	0.000	16/9	0.011	0.000	16/11 6/4	16/9	16/11 6/4	0.000	0.003	16/11 6/4	0.000	0.030	9/7 9/4	0.090	0.080
51	8/7 8/3	0.042	0.034	16/10	0.000	0.000	16/10	16/10	9/8 9/4	0.076	0.075	16/10	0.000	0.000	8/7 8/3	0.042	0.034
44	8/4(2) 7/3(2)	0.000	0.000	8/4(2) 7/3(2)	0.000	0.000	8/4(2) 7/3(2)	8/4(2) 7/3(2)	16/8 7/3(2)	0.076	0.076	8/4(2) 7/3(2)	0.000	0.000	8/4(2) 7/3(2)	0.000	0.000
43	7/3 6/3	0.027	0.028	7/4 7/3	0.000	0.000	7/4 7/3	7/4 7/3	16/12 6/3	0.013	0.014	16/12 6/3	0.013	0.014	7/4 7/3	0.000	0.000
42	16/14 7/3	0.043	0.042	16/12 6/4	0.000	0.004	16/12 6/4	16/10	16/12 6/4	0.000	0.004	16/12 6/4	0.000	0.004	8/6 8/4	0.041	0.019
41	8/7 8/4	0.015	0.000	8/7 8/4	0.015	0.000	7/6 7/3	8/7 8/4	16/11	0.002	0.060	16/11	0.002	0.006	8/7 8/4	0.015	0.000
33	16/10 7/4(2)	0.000	0.000	16/10 7/4(2)	0.000	0.000	16/10 7/4(2)	16/10 7/4(2)	16/13 7/4(2) 6/3	0.023	0.008	16/10 7/4(2)	0.000	0.000	16/10 7/4(2)	0.000	0.000
32	7/4 6/4	0.014	0.060	16/14 6/3	0.002	0.000	16/13 6/4	16/14 6/3	16/13 6/4	0.000	0.003	16/13 6/4	0.000	0.003	16/14 7/4	0.021	0.016
31	16/15 6/3	0.008	0.000	16/15 6/3	0.008	0.000	7/6 7/4	16/15 6/3	7/6 7/4	0.000	0.004	7/6 7/4	0.000	0.004	7/6 7/4	0.000	0.004
22	8/4(2)	0.000	0.000	8/4(2)	0.000	0.000	8/4(2)	8/4(2)	8/4(2)	0.000	0.000	8/4(2)	0.000	0.000	8/4(2)	0.000	0.000
21	6/3	0.004	0.008	6/3	0.040	0.008	16/15 6/4	16/15 6/4	16/13	0.015	0.020	16/15 6/4	0.000	0.000	8/7 8/6	0.067	0.063
11	16/12	0.000	0.000	19/18 16/13	0.006	0.033	16/12	16/12	16/15 10/9 10/8	0.091	0.078	16/12	0.000	0.000	16/15 8/7 8/6	0.049	0.073
TOTAL	Total Error	0.235	0.245		0.247	0.191				0.483	0.554		0.090	0.174		0.579	0.521

Messages In This Thread

• How To Trap An Anchor Bot Comparison Part 1
  Michael Depreli -- Sunday, 2 May 2010, at 2:56 p.m.
  • How To Trap An Anchor Bot Comparison Part 1
    Timothy Chow -- Sunday, 2 May 2010, at 7:12 p.m.
    • How To Trap An Anchor Bot Comparison Part 1
      misja -- Monday, 3 May 2010, at 6:58 a.m.