Working my way through this terrific book, this example made me do a double take. I did rollouts with Gnu on world class, 2-ply, 648 trials...not the final word but certainly points you in the right direction.

I thought the double hit would be best, I was surprised that 13/8 came in second. Not a lot of loss in equity but still #2.

Would love if someone could chime in with a XG roll out.

I can't help with rollouts, but with Hero having
- one home board point vs. two for Villain
- one man back vs. two for Villain

it's certainly reasonable that a passive, "safe" move would be near the top of the list.

13/8 is not really passive. It improves Blacks position well beyond the 5 pips gain by balancing the spare checkers on the 13 and 8 points and bringing a 9th man to strengthen small doublets and other freak blitzes (at least until Red anchors).

It is well known that stacking is almost always wrong, but many early 13/8 plays putting a 4th man on the 8 point (and especially this one) are not stacking plays since they start with an even heavier midpoint.

Regarding the rollout, 648 trials is far from enough to be significant. I don't have a XG rollout but had a 1296 (money, not 3pt match) gnubg one that gives 13/8 and 6/4*/1* too close to call (13/8 marginally ahead) with 13/10 6/4* and 24/21 6/4* unlikely but still possible. 10000 trials would certainly have eliminated the 6/4* plays but may not have been enough to separate the first two choices.

yeah, my first reaction when seeing the position wasn't surprise, that 13/8 did so well, but that the double hit does so well, since it does seem anti-thematic judging by safe vs. bold criterias.

I have XG and its evaluation on its strongest setting match your rollout results (not gonna do rollouts because I have no reason to suspect it won't be just as close).

I use 648 on world class as a compromise on evaluation and brevity. Doing 10000 rollouts is far from practical for me. When you did the 1296 what setting was Gnu set for? on expert (0-ply) it would be pretty quick for 1296. on 3-ply it would probably be around 40mins on my rig.

I realized after I did the rollouts that is was set to match play instead of cash but with the score and 0-0 do you really think it would make a difference?

C'mon, dude. It's a passive move with some slight balancing benefits. But the main point is that it leaves no additional blots. Normally you wouldn't have much motivation at all to put a 4th checker on the 8-point.

1. Rollout¹ 6/4* 4/1* eq:-0.373
Player : 43.05% (G:9.64% B:0.63%)
Opponent: 56.95% (G:18.19% B:1.45%)
Confidence: &plusmn; 0.026 (-0.399<E<-0.347)
Duration: 53 minutes 32 seconds

2. Rollout¹ 13/8 eq:-0.390 (-0.017)
Player : 42.87% (G:7.82% B:0.56%)
Opponent: 57.13% (G:17.63% B:0.68%)
Confidence: &plusmn; 0.019 (-0.409<E<-0.371)
Duration: 46 minutes 12 seconds

3. Rollout¹ 13/10 6/4* eq:-0.427 (-0.054)
Player : 42.30% (G:10.16% B:0.91%)
Opponent: 57.70% (G:20.27% B:2.14%)
Confidence: &plusmn; 0.025 (-0.452<E<-0.402)
Duration: 53 minutes 42 seconds

4. Rollout¹ 24/21 6/4* eq:-0.478 (-0.105)
Player : 41.32% (G:8.68% B:0.70%)
Opponent: 58.68% (G:19.74% B:1.56%)
Confidence: &plusmn; 0.024 (-0.502<E<-0.454)
Duration: 52 minutes 27 seconds


¹ 1296 Games rolled with Variance Reduction.
Dice Seed: 36011002
Moves and cube decisions: 3 ply


eXtreme Gammon Version: 1.21, MET: Rockwell-Kazaross

I just finished a rollout on Snowie (648/2-ply/cube 3-ply) which gave 13/8 as slightly better (.02) than 6/4*/1*.

The only surprise here is that the double-hit does even that well. Black usually needs to be facing some severe concrete tactical threats to make a play like 6/4*/1* a contender. 13/8 looks very correct positionally and I'd make that play over the board in a shot.

^^ IDK, since I already considered the passive play, there's something to be said for the double-hit as well (as the rollouts indicate, of course).

Given that Villain has the 5-point made AND the back men split, Hero isn't real eager to give him complete freedom of movement next roll, because he'll usually find something constructive to do, and any improvement in the forward position in particular is going to be pretty constraining on Hero.

It's not so much that there's a specific, severe tactical threat, but by putting two in the air, Hero hopes to prevent Villain from improving, gain some time, and either scramble his remaining runner to safety or build a forward position.

It apparently just comes down to style/taste/flipping a mental coin.

You can find out very easily. Just open the position you have already rolled out, and click the button that says "MWC." This causes GNUBG to display cash equity, without regard to match standing. Next, click the "Details" button to display the 2-line rollout results. This shows wins, gammons and backgammons for both players. Finally, click and Ctrl+click on the rollout results you are interested in so that they are selected, and then click the "Copy" button. This opens a window showing the details of your rollout in a manner similar to what BigWill has posted from XG. Using standard copy and paste techniques, you can copy them into your next post to this thread!

And, I hope you'll do that. I'm intrigued that your rollout shows a 14% preference, while everyone else is getting only 1-2% difference. I wonder if any of the numbers will change after clicking the "MWC" button.

One difference between cash games and match play that you cannot correct without performing another rollout is the use of the Jacoby Rule. Your rollouts count gammons even when the cube is centered.

Maybe I should be asking for help with GNUBG! Here are the rollout results I got:

Position ID: sOfgASLgc/ADIA
Match ID: cAkNAAAAAAAA

1. Rollout 6/4*/1* Eq.: -0.315
0.430 0.095 0.002 - 0.570 0.180 0.008 CL -0.230 CF -0.315
[0.002 0.002 0.000 - 0.002 0.003 0.001 CL 0.006 CF 0.014]
Full cubeful rollout with var.redn.
648 games, Mersenne Twister dice gen. with seed 900260520 and quasi-random dice
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]

2. Rollout 13/10 6/4* Eq.: -0.362 ( -0.047)
0.424 0.100 0.004 - 0.576 0.201 0.011 CL -0.259 CF -0.362
[0.002 0.002 0.000 - 0.002 0.003 0.001 CL 0.006 CF 0.019]
Full cubeful rollout with var.redn.
648 games, Mersenne Twister dice gen. with seed 900260520 and quasi-random dice
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]

3. Rollout 13/8 Eq.: -0.365 ( -0.051)
0.424 0.072 0.002 - 0.576 0.170 0.003 CL -0.251 CF -0.365
[0.002 0.002 0.000 - 0.002 0.003 0.001 CL 0.005 CF 0.013]
Full cubeful rollout with var.redn.
648 games, Mersenne Twister dice gen. with seed 900260520 and quasi-random dice
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]

4. Rollout 24/21 6/4* Eq.: -0.392 ( -0.077)
0.413 0.085 0.004 - 0.587 0.185 0.008 CL -0.278 CF -0.392
[0.002 0.002 0.001 - 0.002 0.003 0.001 CL 0.006 CF 0.018]
Full cubeful rollout with var.redn.
648 games, Mersenne Twister dice gen. with seed 900260520 and quasi-random dice
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]

I think I have my rollouts set for cubeless evaluations, maybe that's why? What's the best settings for rollouts?

I think when I find myself in this position I'm going to use Bill's advice...just because Gnu says it's 2nd best doesn't mean that Gnu's right...

Your settings are fine, you are just using mwc (match winning chances) instead of emg (equivalent to moneygame) equity. Also I think Tapir_Mike misread your rollout. The difference was .14% not 14%.

I would recommend using emg instead, though. With mwc gnu is gonna tell you that pretty much any mistake you make early in long match is a small one (because winning or losing the first game in a 15 point match doesn't have a very large impact on your overall match winning chances), but make the same mistake at double matchpoint and it is now a huge blunder.

Aha! Thank you, Mute.

So, what's up with ranking I got? Any ideas?

I rolled the d*** thing out for over five hours!

Why exactly concerns you about your rollout results?

24/21 13/11 .... nuts to equities!

It bothers me that mine the only result that shows 13/8 in third place, 5.1% behind! Bill Robertie had 13/8 first (in snowie), and Nick had it only 0.14% behind (albeit with MWC misset).

When different bots give different rollout results, it's time to stop and actually look at the position. That 13/8 might be the correct play makes sense to me: it balances Black's builders and avoids leaving blots when White has a stronger board and Black has already escaped a back checker.

If someone can construct a good argument why 6/4*/1* is best in this sort of position, I've yet to hear it.

Apparently my argument above, that it stops Villain from making progress in a position where the vast majority of his rolls are constructive, isn't "good". Oh well.

The other potential issue here is that Hero is in imminent danger of facing a cube turn. In a -3:-3 spot (as in the OP), where the trailer is significantly more reluctant to take an initial cube than in a cash game, that could have a big impact on the outcome, because the cube turn is much more likely to be efficient. In that case, Hero should lean towards the play with higher volatility, because he will get the upside with no corresponding downside, and it seems clear that 6/4*/1* is more volatile.

It's not clear to me whether all the rollouts in this thread were 3-point matches or not, but that could have something to do with it.

If that's not the issue, then who knows. When Snowie plays Snowie, it likes 13/8. When gnu plays gnu, it likes 6/4*/1*. I don't play as well as either, so I'm not all that worried about it.

GNU's 1296 3-ply rollout actually favors 13/8 by .017

Thanks for this. Would you mind clicking the Copy button, and posting the rollout data here?

- Mike

Unfortunately I just looked at the results and closed the program, and since it takes about 1.5 hr on my computer I can't reproduce it on the fly. Will try to run it again later in the week.

No problem. I was hoping you might have saved it. There's no need to rerun it.

My rollout used "World Class," 2-ply settings. The difference was most likely that you ran 3-ply tests.

C'ya,

- Mike