Game 1, move 3: forum take decision

[Zorba]

4HPwA0AzTuEBKA:MBmgAAAAAAAA

[diane]

Spoiler

I am basing my decision to take as much on this post, as because I think it is a take.  http://www.fibsboard.com/feedback-questions/playing-backgammon-against-the-computer-t2350.0.html;msg17560#msg17560  Unless I have misunderstood this post, this is saying that with one on the bar, on roll, roadkill has the double, and we should take.

[close]

[stog]

Spoiler

if this was a 7 or 9 point i would take but it is a 5 pointer and we cant afford to go 4 down so....

[close]

[ah_clem]

Spoiler

His assets: a three point board, two builders threatening to make a fourth point (9/36 or 25% chance to do this next roll), an opponent's checker on the roof, and a 15 point lead in the pipcount.  Thus a lead in position, race, and threats - definitely a double.

Our assets: one escaped backman. That's it.   We're in bad shape here. PASS.

[close]

[NIHILIST]

Spoiler

Remember we talked about THE THREE Ps ? Pips, prime, position.................how many of those 3 do we have ?


PASS.

[close]

Bob

[Zorba]

Spoiler

I think we can eek out a take here. An early 5-5 blitz double happens quite often, so it's handy to have reference positions for it. My closest reference here is if we ignore road's opening 4-1, and pretend we opened with a 6-5, followed by this 5-5 hit and dance. That position is in fact not good enough to double, although it's close. It's a huge take then, obviously.

Here, road has several advantages compared to that one: 5 pips gained in the race (a minor consideration), split backcheckers (this may look unimportant, but it can greatly help when the blitz fails. It increases anchoring chances then, but also chances to run home ASAP and it creates extra flexibility, which is very valuable after a failed blitz), and most important of all, a builder on the 9, which greatly increases chances to make another homeboard point, or attack us if we enter.

These improvements, especially that extra builder, make it a very strong double. But it seems unlikely that this would be enough to turn a no double into a double/pass. roadkill has no other blot to shoot at, as is the case with many other 5-5 blitz doubles.

The matchscore makes doubled gammons more valuable than usual and this surely is a very gammonish double. However, the matchscore also makes our recubes more powerful and that compensates for a large part. It's probably a bigger take for money, but not a large difference.

[close]

[blitzxz]

Spoiler

We are behind in race. We have horrible inflexible position. We are on bar and roadkillbooks is threatning immediate close out with three builders and more to come. And doubled gammons lead to valuable crawford game. PASS.

[close]

[diane]

Spoiler

Man, you lot are CHICKEN     I stick with take, but looks like I will be outvoted, unless Zorba made a persuasive argument there  

[close]

[ah_clem]

Spoiler

I think we can eek out a take here.

When I first looked at the position I thought it was a take too.  Surely we have more than a 25% chance of coming back to win.  But then I looked at the looming blitz, and thought about how often we get gammoned here.  I don't think it's worth the risk of being 4 down in a match to 5.

My hunch is that it's a fairly close take/pass decision, but I'll stand by my decision to bail now before the whole match goes south.

[close]

[lewscannon]

Spoiler

I vote pass. This reminds me of some horrible, quick matches against bots. Roadkill has some excellent chances to make another point in his home board. We have nothing particularly good happening anywhere on the board. Run away and live to fight another day.

[close]

[playBunny]

Cube Rollout: 1296 games, 2-ply throughout

Spoiler

It's no blunder to have dropped.
 
Cubeless equity  +0.590 (money:  +0.572)
 

1 .. Double, take ..... 0.960
 
2 .. Double, pass ..... 1.000 .. (+0.040)
 
3 .. No double ........ 0.804 .. (-0.156)
 
Proper cube action: Double, Take

[close]

[diane]

    I love being right, I also felt I had to mention it, since it happens so rarely 

I told you you lot were chickens 

[roadkillbooks]

So there is a huge difference between that roll out on Snowie.  Snowie  shows the results as follows:

1.1 32.8 63.1 36.9 7.8 .4

1. Double take .773
2. No double .715 (-0.057)
3. DOUBLE, PASS 1.00 (+ .227)

Snowie says that this is a blunder of  a pass decision.  In all actuality...I would take or pass this depending on who I was playing.  Essentailly someone that was quite weak AND agressive I would probally drop.  They would be inclined to be able to pull off the blitz better for there aggressiveness.  It would probally just be easier to battle for 0-1 than risk going down.  I was schooled from Woolsey's Backgammon Encyclopedia's reference positions and Zorba's comment was right on the money.  If this were my first move (without the extra buidler) I wouldnt yet have a double. Here there is only one piece out.

To correct one comment..I didnt want to say anything while the poll was open.  I have 15 point number not including 6-4 which I would still make the 2 point with. 3-1,4-1,4-2, 5-2,4-3,5-4,2-2,4-4 and 6-4

I had no idea how this was going as I am blinded and cant see the vote.

thx
roadkilbooks

[vikingblood80]

I heard that Snowie does not take match-scores into account?

So the argument of trailing 0:4* as concerning the match score is the main issue to possibly drop this. Anyway I also would have taken it. 

[blitzxz]

I also made gnu rollout. 2-ply with 5000 games. Borderline and not statitically significant on 5-procent level.

Rollout details:
Centered 1-cube:
  0,650 0,335 0,005 - 0,350 0,066 0,004 CL  +0,590 CF  +0,807
[0,001 0,001 0,000 - 0,001 0,001 0,000 CL   0,002 CF   0,004]
Player Forum owns 2-cube:
  0,657 0,344 0,004 - 0,343 0,064 0,005 CL  +1,323 CF  +0,992
[0,001 0,001 0,000 - 0,001 0,001 0,001 CL   0,005 CF   0,007]
Full cubeful rollout with var.redn.
5000 games, Mersenne Twister dice gen. with seed 132562243 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]

[roadkillbooks]

I heard that Snowie does not take match-scores into account?

So the argument of trailing 0:4* as concerning the match score is the main issue to possibly drop this. Anyway I also would have taken it. 

OK...I didnt plunk the  match score into snowie...it was human error.  The result of a rollout with the match set at 5..is less.. an error of .084

Snowie does take it into account though...Image if the score was me 3 forum 1....

anyway sorry about the mishap

[Zorba]

And yet another rollout, same settings as blitzxz but slightly smaller 2-ply filter (0.12 instead of 0.16). This gives a borderline take >95% confidence.

For bot aficionados: GnuBG 2-ply and 4-ply evaluations both got around a 0.92 take, so this fits into the pattern of GnuBG underestimating blitzer's chances. GnuBG 3-ply got it as a small drop, 1.02.

I also rolled this cube decision out at lower settings and it leads to a bigger take (0.95, 0.93).

I wonder about Snowie's values, it wins a lot less games and gammons for the blitzer. What settings did Snowie play at in the rollout and what are the confidence intervals?

There's also the MET issue to consider, sometimes the Snowie MET and GnuBG's default g11 MET lead to quite different results, even if the w/g/bg breakdown is similar.


Rollout details:
Centered 1-cube:
   65,05  33,51   0,61 -  34,95   6,56   0,38 CL  +0,5934 CF  +0,8113
[  0,07   0,11   0,05 -   0,07   0,06   0,04 CL   0,0021 CF   0,0048]
Player FIBSBoard forum owns 2-cube:
   65,61  34,36   0,52 -  34,39   6,53   0,69 CL  +1,3167 CF  +0,9810
[  0,09   0,17   0,07 -   0,09   0,08   0,06 CL   0,0054 CF   0,0073]
Full cubeful rollout with var.redn.
3888 games, Mersenne Twister dice gen. with seed 859211562 and quasi-random dice
Play:  2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 8 more moves within equity 0,12
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]