Position # 57

[PersianLord]

Position information:

Pip-count:  Red 78-91 White
Score:       Red   2-2  White
Match:               3-pointer

Red to play 5-2

Spoiler

This position was sent to my e-mail by NIHI. As of his suggestion, I didn't set up a poll. We want to know how the FIBSters think and probe this position. I am sure we will get a good laugh, buwahahahahahahaha

[close]

[ah_clem]

Spoiler

Safe play or bold play?

24/17 (safe)
or
17/15* 10 (bold)

Favoring safe play:
Gammons don't matter, so there's nothing to gain by going for gammon.
After the move red will be 20 pips ahead in the race.
When ahead, simplify.
Red has a nice distribution, where white doesn't. (and with all those checkers on the 18 white will have little choice in how he enters, so he may get a good distribution, or he may not...)

Favoring bold play:
Red has more checkers back than white
Red has a 5 point board vs white's two point board.

So, there are some arguments in favor of both.

My take?  Breaking contact and transitioning to a race with a 20 pip lead at DMP is enough of a sure thing to sway me in that direction.  Not sure I'm right, but there's enough to go wrong with the bold play to dissuade me. 

If gammons were in play, I'd definitely hit without thinking about it too hard.

[close]

[dropper]

Spoiler

24/17 as there's a clear race lead. Even if gammons were counting, I don't see tons of them happening with a partially crunched board. Staying back risks losing ground by rolling high doubles or other non-escaping rolls.

I think it'd be a different call if our board were better or at a go-gammon score.

[close]

[diane]

Spoiler

24-17, the hit is just introducing more to go wrong with a nice looking race. It is one of those I probably don't stop and think about at the time though, and might well do then regret

[close]

[ah_clem]

So, do we get a rollout/answer here? 

[blitzxz]

Spoiler

I have a some kind of rule in these but I don't know if it is correct. If you're leading in pips more then take (over 12 % lead) play safe, if not hit. But I adjust it to position. Here red is behind prime, far from home and white's distribution makes lead bigger so safe play is propably good even with smaller pip lead but if gammons would matter then that would be another thing again. White has a big pile of checkers outside homefield so there could be quite many gammons. So I would normally hit (although I would have bad feeling about it) but in this score safe play is very clear.

[close]

[dorbel]

Like so many bg positions, the answer to "Should I hit if gammons count?" is, "It depends". At 2-away, crawford, should you hit? No, hitting is an error and 24/17 is correct. That play will win about 80% of the time and you may even fluke a gammon (2-3%) if White rolls 2-2 or 1-1 next and you hit. If you hit now on the other hand, your gammons will go up to about 10%, but this is not enough to compensate for the fact that your overall wins will drop to something closer to 70%.
How about 4-away, 2-away? If the cube is centred, then you should hit. If you have already doubled, then the safe 23/17 is correct! What is going on here? With the cube in the middle, you get a big cash when White dances after you hit, but if White owns the cube you won't be able to do that and you will actually have to win the game by playing it until the end. There just aren't enough gammons in this position to compensate for the fact that Red has to escape very quickly after hitting or crash.
The key to understanding this position is that Red's board can only get worse, whereas White's can only get better! 24/17 is always right with the sole exception of the 4-away, 2-away centre cube position, where hit and (hope to) cash is the order of the day. Note that at most normal scores you can double (and get a take) after the incorrect hit and a dance, but you usually have a much stronger double after making the safe play anyway.
The figures quoted here are from relatively short snowie rollouts. Gnu may differ a little and IMO will probably be a little more accurate, but the underlying principles should remain the same.

[Zorba]

Here's the GNUBG evaluations and quick and dirty 0-ply rollout result. Whopper to hit at DoubleMatchPoint.

    1. Cubeful 2-ply    24/17                        Eq.:  +0,5605
        78,02   2,40   0,00 -  21,98   0,29   0,00
        2-ply cubeful prune [world class]
    2. Cubeful 2-ply    17/15*/10                    Eq.:  +0,3994 ( -0,1611)
        69,97  10,04   0,02 -  30,03   2,19   0,01
        2-ply cubeful prune [world class]

    1. Cubeful 3-ply    24/17                        Eq.:  +0,5764
        78,82   1,81   0,01 -  21,18   0,40   0,00
        3-ply cubeful prune [grandmaster]
    2. Cubeful 3-ply    17/15*/10                    Eq.:  +0,4628 ( -0,1135)
        73,14   7,33   0,02 -  26,86   1,78   0,01
        3-ply cubeful prune [grandmaster]

    1. Cubeful 4-ply    24/17                        Eq.:  +0,5828
        79,14   2,47   0,00 -  20,86   0,35   0,00
        4-ply cubeful prune
    2. Cubeful 4-ply    17/15*/10                    Eq.:  +0,3987 ( -0,1841)
        69,94  11,01   0,02 -  30,06   2,13   0,01
        4-ply cubeful prune
============================================
    1. Rollout          24/17                        Eq.:  +0,5859
        79,29   4,83   0,01 -  20,71   0,48   0,00 CL  +0,5859
      [  0,12   0,38   0,01 -   0,12   0,08   0,07 CL   0,0024]
        Full cubeless rollout (trunc. at one-sided bearoff) with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 818332620 and quasi-random dice
        Play: 0-ply cubeful prune
        Cube: 0-ply cubeful prune
    2. Rollout          17/15*/10                    Eq.:  +0,4464 ( -0,1395)
        72,32  18,38   0,05 -  27,68   2,29   0,05 CL  +0,4464
      [  0,22   0,70   0,02 -   0,22   0,19   0,03 CL   0,0045]
        Full cubeless rollout (trunc. at one-sided bearoff) with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 818332620 and quasi-random dice
        Play: 0-ply cubeful prune
        Cube: 0-ply cubeful prune

[Zorba]

How about 4-away, 2-away? If the cube is centred, then you should hit. If you have already doubled, then the safe 23/17 is correct!

I checked it with GNUBG because this seemed very unlikely to me. I think you might've misinterpreted something. It's very wrong to hit when you still have to double. Maybe even more so at 4-away 2-away. When the hit works, it's overkill, because you're gonna cash instead of winning those extra gammons, so you just get to -3,-2 and the gammons are useless. When opponent enters on the sixpoint (11 rolls), you don't have much cube leverage anymore, if at all. 6-4 and 6-6 are horror returns after which you become an instant clear underdog.

After the quiet 24/17, you have a lot of decent to highly efficient cubes (small drops or small takes), simply because of the race advantage. Only the doubles from 3-3 up (4 rolls) give your opponent decent winning chances and a good take or even no double for you.

Now, at -4,-2 with opponent holding a 2-cube, gammon wins are twice as valuable here because you need exactly four points to win the match. So that's a much better time to hit. If hitting wins 10% extra gammons, at the cost of losing 10% more single games, then at the pure Gammon-Go scores of -2,-1Cr and -4,-2 (opp owns 2-cube) the plays would be equal.

As the evaluation gets win and gammon percentages very close to that, a rollout is needed for those.

[NIHILIST]

Isn't anyone going to tell me how brilliant I was for playing 24-17 in the actual game ?

Bob

[dorbel]

Absolutely right Zorba. 24/17 is correct at all times, for the reasons that you give, particularly the very inefficient cube after the dance. I fell victim to a Snowie bug, when for reasons of its own it looks at that position at 4-away, 2-away and evaluates it at 2-ply and gets it badly wrong. Well spotted.

[Zorba]

It looks like the hit may just be right at pure gammon-go scores, where gammons are twice as valuable as usual. F.i. trailing -4,-2 with the cube already given, or trailing -2,-1 in the Crawford game: any loss loses the match, a single win makes the match equal, a gammon win wins the match:

    1. Rollout          17/15*/10                    Eq.:  +0,5952
        72,36   7,40   0,11 -  27,64   2,62   0,05 CL  +0,5952 CF  +0,5952
      [  0,10   0,21   0,04 -   0,10   0,23   0,02 CL   0,0049 CF   0,0049]
        Full cubeful rollout with var.redn.
        2592 games, Mersenne Twister dice gen. with seed 818332620 and quasi-random dice
        Play:  2-ply cubeful prune [world class]
        keep the first 0 0-ply moves and up to 12 more moves within equity 0,12
        Skip pruning for 1-ply moves.
        Cube: 0-ply cubeful prune [expert]
    2. Rollout          24/17                        Eq.:  +0,5723 ( -0,0229)
        78,08   0,53   0,02 -  21,92   0,35   0,02 CL  +0,5723 CF  +0,5723
      [  0,06   0,06   0,01 -   0,06   0,15   0,02 CL   0,0018 CF   0,0018]
        Full cubeful rollout with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 818332620 and quasi-random dice
        Play:  2-ply cubeful prune [world class]
        keep the first 0 0-ply moves and up to 12 more moves within equity 0,12
        Skip pruning for 1-ply moves.
        Cube: 0-ply cubeful prune [expert]


Here, the hit loses 5.7% more games overall, but gains an extra 6.9% gammons. At Gammon-Go scores with gammon value (price) of 1, that makes the gammonish play correct.

Now if Bob (or anybody else) had gotten that one right, I might've considered it brilliant!

P.S. Tom Keith' site has a very good glossary of backgammon terms, for those of you who may wonder about some of the terms being used:
http://bkgm.com/glossary.html

[PersianLord]

Isn't anyone going to tell me how brilliant I was for playing 24-17 in the actual game ?

Bob

Glory be to thee, leviathan.

PL

[NIHILIST]

Finally some respect.

Bob