|
| From: | Mary Hickey |
| Subject: | RE: [Bug-gnubg] Too good to double when a gammon is impossible?! |
| Date: | Fri, 21 Aug 2009 20:27:33 -0400 |
|
Hi Tim, This happens with Snowie too. If you are a member of GammonVillage, you can find Problem C1 in my column titled "Playing with Fire". The rollout for it using evaluations correctly called it a double/pass, but the Live Cube messed up by, as Christian notes is probably what's going on with the one you sent, getting later cube actions wrong. Apparently there were enough of them to swing its decision to "too good" even though of course it could not be...the opponent had 11 checkers off! Ian Shaw asked what was up with that, and here is what I responded then: Your question regarding Snowie's too-good evaluation is interesting. I've seen this before, and what I believe is that Snowie doesn't look at gammons per se when it calculates this. Instead, it looks at how much it would win, including the effects of cube turns, were it to hold instead of double, and compares that to how much it would win were it to double now. Snowie substantially misjudged this position on 3-ply, saying it was a .481 no-double/take, so may very well have also misjudged some of the subsequent positions that arose from it. In this case, since 3-ply is far too optimistic regarding the taker's chances, these later misjudgments in the live-cube analysis would tend to be mostly later takes that should have been passes. After the rollouts were complete, these bad takes would make the equity greater for Black if he waited to redouble. File: C:\Snowie Documents\My exported files\GV 29 PlayingWithFire C1 WithNumbers.txt, 8:18:12 PM, 8/21/2009 Snowie4 Professional Edition Version 4.7.1 Output (Export v2.10) -------------------------------------------------------------------- | GammonVillage 29 Problem C1 Mar 10 2004 | | You (X) vs. Balrog (O) | -------------------------------------------------------------------- Money session. Score X-O: 4-13 O on roll, cube action +-1--2--3--4--5--6--------7--8--9-10-11-12-+ | O O O O O O | | | | O O O O O | | | | O O O | | | S | | X | | n | | X | | o | |BAR| | w | | | | i | | | | e | | | | | X | | | | O X | | | +24-23-22-21-20-19-------18-17-16-15-14-13-+ Pipcount X: 54 O: 80 X-O: 4-13/Money (1) Men Off X: 11 O: 0 CubeValue: 8, O owns Cube Rollout Money equity: 0.580 0.0% 0.0% 79.0% 21.0% 0.1% 0.0% 95% confidence interval: - money cubeless eq.: 0.580 ±0.007, - live cube no redouble: 1.034 ±0.018, - live cube redouble take: 1.028 ±0.019. Rollout settings: Full rollout, 1440 games (equiv. 47002 games), played 2-ply (4), cube 3-ply, settlement 0.550 at 8 pts, random seed, without race database. Evaluations 1. Redouble, pass 1.000 2. No redouble 0.956 (-0.044) 3. Redouble, take 1.007 (+0.007) Proper cube action: Redouble, pass Live cube 1. No redouble 1.034 2. Redouble, pass 1.000 (-0.034) 3. Redouble, take 1.028 (-0.006) Proper cube action: Too good to redouble, pass - ------------------------------ End ---------------------------------- > From: address@hidden > To: address@hidden > Subject: [Bug-gnubg] Too good to double when a gammon is impossible?! > > I just rolled out a cube decision and gnubg evaluated "no double" at > greater than +1.000 equity despite the fact that a gammon was no longer > possible. Below is the text output; I can email the .sgf file if anyone > wants it. > > Tim > > > The score (after 0 games) is: gnubg 0, tchow 0 > Match Information: > > Date: 08/21/2009 > Move number 1: gnubg on roll, cube decision? > > GNU Backgammon Position ID: bgAAULZtEwAiAA > Match ID : AQEAAAAAAAAA > +12-11-10--9--8--7-------6--5--4--3--2--1-+ O: gnubg (Cube: 2) > | O O | | O O O O O X | 0 points > | O | | O O O O O | On roll > | | | | > | | | | > | | | | > ^| |BAR| | > | | | | X > | | | | X > | | | X | XX > | | | X X | XX > | | X | O X X O | XX 0 points > +13-14-15-16-17-18------19-20-21-22-23-24-+ X: tchow > Pip counts: O 108, X 61 > > * gnubg doubles > > Alert: wrong double ( -0.167)! > Rollout details: > Player gnubg owns 2-cube: > 0.828 0.000 0.000 - 0.172 0.007 0.001 CL +0.647 CF +1.167 > [0.002 0.000 0.000 - 0.002 0.000 0.000 CL 0.004 CF 0.010] > Player tchow owns 4-cube: > 0.859 0.000 0.000 - 0.141 0.006 0.001 CL +1.422 CF +1.325 > [0.002 0.000 0.000 - 0.002 0.000 0.000 CL 0.008 CF 0.010] > Full cubeful rollout with var.redn. > 1296 games, Mersenne Twister dice gen. with seed 866958217 and > quasi-random dice > Play: 0-ply cubeful [expert] > Cube: 0-ply cubeful [expert] > > > > _______________________________________________ > Bug-gnubg mailing list > address@hidden > http://lists.gnu.org/mailman/listinfo/bug-gnubg |
| [Prev in Thread] | Current Thread | [Next in Thread] |