Two Plus Two Forums
Sports and Games
Chess and Other Board Games
Backgammon Forum hosted by Bill Robertie.
obvious is not always right
02-17-2015
, 10:01 AM
Doing a 2-point match the outcome turns out to be not very much different. But when the blot is removed to white's 13-point, well, you are right...
Code:
1. 13/7 6/3(2) Eq.: +0,828
0,779 0,383 0,006 - 0,221 0,041 0,002 CL +0,906 CF +0,828
2. 6/3(2) 4/1*(2) Eq.: +0,818 ( -0,010)
0,750 0,450 0,002 - 0,250 0,042 0,002 CL +0,909 CF +0,818
Truncated cubeful rollout (depth 10) with var.redn.
147 games, Mersenne Twister dice gen. with seed 767934765 and quasi-random dice
Stop when std.errs. are small enough: ratio 0,1 (min. 144 games)
Play: world class 2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 8 more moves within equity 0,16
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]
02-18-2015
, 08:54 AM
Position ID: sr0jAAaLm8EBMA Match ID: cAkHAAAAAAAE
02-18-2015
, 02:36 PM
02-18-2015
, 03:52 PM
At the risk of driving Fllecha from the game 
, I would actually play 24/23*/17. It's only a few more shots and it appears to better when missed and better when hit (on average).
Or another way to look at it is -- getting the checker over the prime looks like a much more important thing than "upgrading" from the 8 to the 2 point.
The duplication of 2s after jumping over the prime is real, but I would still jump even if the White checker was moved from the 10 to the 11 point, though it's a closer call.
, I would actually play 24/23*/17. It's only a few more shots and it appears to better when missed and better when hit (on average). Or another way to look at it is -- getting the checker over the prime looks like a much more important thing than "upgrading" from the 8 to the 2 point.
The duplication of 2s after jumping over the prime is real, but I would still jump even if the White checker was moved from the 10 to the 11 point, though it's a closer call.
02-19-2015
, 12:27 AM
02-19-2015
, 04:13 AM
What relieves me is the fact that with out duplication the equity between moves is relatively close so as yoghi pointed out here duplication helps a lot.
02-20-2015
, 01:01 AM
Position ID: sr0jAAaLm8EBMA Match ID: cAkHAAAAAAAE
White - Pips 117
Black - Pips 134
Black to Play 6-1
I made three rollouts in XG to check how the position of White's blot affects the margin by which play R = 24/23*/17 beats play C = 24/23* 8/2. The results are not too far from what Yogiman got using GnuBg. This is not surprising. GnuBg plays a bit worse than XG if you compare its 2-ply World Class mode against XG's 3-ply/XGR setting. The difference is very small. Making a rollout brings the bots into even closer accord.
Yogiman's rollouts use good settings. His standard is a 10-ply (i.e., depth) truncated rollout made in World Class mode. It's a good compromise between accuracy and speed that will give decent results in most "normal" positions. For complex backgames, containment games, and other known "problem" positions, you might need something stronger. For most of the positions in this thread, however, including Problem 41, I would expect Yogiman's GnuBg rollouts to be just as good as XG rollouts. The only thing I would change is to increase the minimum number of trials to, say, 324, or preferably, 648. This is especially important if you are trying to estimate the margins between plays.
In the original position, Yogiman's GnuBg rollout puts the margin between plays R and C at 159 millipoints (i.e., R beats C by 0.159). My 5k XG rollout gives 148. Moving the opponent's blot back to the 14pt produced a margin of 57 millipoints in the 147-trial GnuBg trunc. My 5k XG rollout puts it at 63. Moving the blot forward to the 16pt results in a margin of 26 millipoints in GnuBg. My 31k XG rollout yields 15. The differences here are not primarily differences between GnuBg and XG. They are differences caused by the disparity in the number of trials.
In order to test the effect of duplication, I like the variant with the blot pushed back to the 14pt. Without duplication, the margin between plays R and C falls by 148 – 63 = 85 millipoints. I think this is the true measure of the value of duplication in the OP.
The variant with the blot moved forward, however, creates new attacking options for the opponent. It gives White an extra builder targeting the open 3pt in his board. For this reason, I discount this variant. I don't think it is of much value in determining the effect of duplication. Yogiman points this out in his post above.
Focusing on the variant with the blot pushed back to the 14pt, we see that jumping out with play R is still 63 millipoints better than covering with play C. That's after duplication has been eliminated. To my way of thinking, this demonstrates that escaping the back checkers and keeping a strong presence in the outfield, rather than duplication, is the most important factor in Problem 41. 63 millipoints may not be a blunder, but it does not fit my definition of "relatively close" either.
Bill puts quite succinctly, "In jump-or-cover decisions, jumping is correct more often than not."
Mike
White - Pips 117
Black - Pips 134
Black to Play 6-1
I made three rollouts in XG to check how the position of White's blot affects the margin by which play R = 24/23*/17 beats play C = 24/23* 8/2. The results are not too far from what Yogiman got using GnuBg. This is not surprising. GnuBg plays a bit worse than XG if you compare its 2-ply World Class mode against XG's 3-ply/XGR setting. The difference is very small. Making a rollout brings the bots into even closer accord.
Yogiman's rollouts use good settings. His standard is a 10-ply (i.e., depth) truncated rollout made in World Class mode. It's a good compromise between accuracy and speed that will give decent results in most "normal" positions. For complex backgames, containment games, and other known "problem" positions, you might need something stronger. For most of the positions in this thread, however, including Problem 41, I would expect Yogiman's GnuBg rollouts to be just as good as XG rollouts. The only thing I would change is to increase the minimum number of trials to, say, 324, or preferably, 648. This is especially important if you are trying to estimate the margins between plays.
In the original position, Yogiman's GnuBg rollout puts the margin between plays R and C at 159 millipoints (i.e., R beats C by 0.159). My 5k XG rollout gives 148. Moving the opponent's blot back to the 14pt produced a margin of 57 millipoints in the 147-trial GnuBg trunc. My 5k XG rollout puts it at 63. Moving the blot forward to the 16pt results in a margin of 26 millipoints in GnuBg. My 31k XG rollout yields 15. The differences here are not primarily differences between GnuBg and XG. They are differences caused by the disparity in the number of trials.
In order to test the effect of duplication, I like the variant with the blot pushed back to the 14pt. Without duplication, the margin between plays R and C falls by 148 – 63 = 85 millipoints. I think this is the true measure of the value of duplication in the OP.
The variant with the blot moved forward, however, creates new attacking options for the opponent. It gives White an extra builder targeting the open 3pt in his board. For this reason, I discount this variant. I don't think it is of much value in determining the effect of duplication. Yogiman points this out in his post above.
Focusing on the variant with the blot pushed back to the 14pt, we see that jumping out with play R is still 63 millipoints better than covering with play C. That's after duplication has been eliminated. To my way of thinking, this demonstrates that escaping the back checkers and keeping a strong presence in the outfield, rather than duplication, is the most important factor in Problem 41. 63 millipoints may not be a blunder, but it does not fit my definition of "relatively close" either.
Bill puts quite succinctly, "In jump-or-cover decisions, jumping is correct more often than not."
Mike
Last edited by Taper_Mike; 02-20-2015 at 01:13 AM.