[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Cube eval going crazy on this one
From: |
Christian Anthon |
Subject: |
Re: [Bug-gnubg] Cube eval going crazy on this one |
Date: |
Tue, 10 Mar 2009 14:15:21 +0100 |
I haven't looked at the thread, but the net is rather bad in handling
late hits. Call it odd/even ply if you like, but it is just something
the net needs to be trained in, or a suitable sub-net should be
constructed. Anyway doing a cubeless rollout is better when the net is
bad, as the play is at least more reasonable, and you don't get the
win/loss fractions messed up by truncation using a bad net.
Christian.
On Thu, Feb 26, 2009 at 10:23 AM, Massimiliano Maini
<address@hidden> wrote:
>
> Hi all,
>
> on bgonline this cube decision has been posted and the evals returned by
> gnubg
> are crazy on even ply:
>
> GNU Backgammon Position ID: XgAAILftgAARAA
> Match ID : AQEAAAAAAAAA
> +-1--2--3--4--5--6-------7--8--9-10-11-12-+ O: gnubg (Cube: 2)
> | O O O O O | | | 0 points
> | O O O O O | | | On roll
> | O O | | |
> | | | |
> | | | |
> | |BAR| |^
> X | | | |
> XX | X | | |
> XX | X | | |
> XX | X | | |
> XX | O X X O | X | O | 0 points
> +24-23-22-21-20-19------18-17-16-15-14-13-+ X: MaX
>
> The original one was in ma match, with score 9aw-9aw: I changed it to money
> to simplify since it doesn't make any difference.
>
> Cube analysis
> 0-ply cubeless equity +0.190
> 0.610 0.000 0.000 - 0.390 0.027 0.003
> Cubeful equities:
> 1. No double +0.398
> 2. Double, pass +1.000 ( +0.602)
> 3. Double, take +0.116 ( -0.282)
> Proper cube action: No redouble, take (31.9%)
>
> Cube analysis
> 1-ply cubeless equity +0.464
> 0.747 0.000 0.000 - 0.253 0.027 0.003
> Cubeful equities:
> 1. Double, take +0.756
> 2. Double, pass +1.000 ( +0.244)
> 3. No double +0.695 ( -0.062)
> Proper cube action: Redouble, take
>
> Cube analysis
> 2-ply cubeless equity +0.245
> 0.637 0.000 0.000 - 0.363 0.026 0.003
> Cubeful equities:
> 1. No double +0.462
> 2. Double, pass +1.000 ( +0.538)
> 3. Double, take +0.244 ( -0.218)
> Proper cube action: No redouble, take (28.9%)
>
> Cube analysis
> 3-ply cubeless equity +0.503
> 0.765 0.000 0.000 - 0.235 0.025 0.003
> Cubeful equities:
> 1. Double, take +0.844
> 2. Double, pass +1.000 ( +0.156)
> 3. No double +0.796 ( -0.048)
> Proper cube action: Redouble, take
>
> Cube analysis
> 4-ply cubeless equity +0.288
> 0.658 0.000 0.000 - 0.342 0.025 0.003
> Cubeful equities:
> 1. No double +0.532
> 2. Double, pass +1.000 ( +0.468)
> 3. Double, take +0.342 ( -0.189)
> Proper cube action: No redouble, take (28.8%)
>
> Position ID: XgAAILftgAARAA
> Match ID: AQEAAAAAAAAA
> Evaluator: Crashed
> Win W(g) W(bg) L(g) L(bg) Equity Cubeful
> static: 0.610 0.000 0.000 0.027 0.003 +0.190 +0.398
> 1 ply: 0.747 0.000 0.000 0.027 0.003 +0.464 +0.695
> 2 ply: 0.637 0.000 0.000 0.026 0.003 +0.245 +0.462
> 3 ply: 0.765 0.000 0.000 0.025 0.003 +0.503 +0.796
> 4 ply: 0.658 0.000 0.000 0.025 0.003 +0.288 +0.532
>
>
> Cubeless winning chances swing a lot between even and odd plies, while
> gammon and backgammon percentages are very stable.
> Notice that even W% increase 0/2/4ply, but so do odd ones 1ply/3ply.
>
> Can we still blame this on odd-even effect ?
>
> MaX.
>
> P.S.
> The thread on bgonline is this one:
>
> http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=35362
>
> Neil said that it wasn't worth to do a 2ply cube rollout since the results
> would be garbage, due to the fact 2ply eval screws this position big time.
> However, the (stopped) rollout 2ply chequer/2ply cube gave the same result
> as the (stopped) rollout 2ply chequer/3ply cube and both were similar to
> 3ply
> eval. Rollout/eval figures are in this post:
>
> http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=35416
>
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>
>
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- Re: [Bug-gnubg] Cube eval going crazy on this one,
Christian Anthon <=