Re: Requested gnubgautorcOne thread or four is making no difference at all 
here. Each just gets the same result it got before. There is a sort of 
consistency in that the wrong result can be reproduced just like the right one. 
I just don't know exactly what triggers a move between them. Mike can you get 
3-ply up as showing No double/Take? If you do I bet a rollout from there will 
be very close to my initial .847 take rather than the correct .877 take.
  ----- Original Message ----- 
  From: Michael Petch 
  To: Neil Robins 
  Cc: [email protected] 
  Sent: Friday, August 28, 2009 9:30 AM
  Subject: Re: Requested gnubgautorc


  I think this was asked earlier by someone  else. Have you tried both these 
versions of Gnubg where Threads have been set to one on each. One thing some of 
us have noticed is that results can vary based on threads being higher than 1. 
I am just curious. What results on rollouts do you get when you run with 1 
thread only (I am trying to eliminate the most obvious reason for a possible 
difference). If I understand your output below you have 2 versions with same 
settings producing different results.


  On 28/08/09 2:15 AM, "Neil Robins" <[email protected]> wrote:


    TRUMP THIS FOR NONSENSE

    I now have two versions of GNU open on my computer and after a lot of trial 
and error, clicking around etc, but changing no basic settings, I seem to have 
achieved getting one each to reproduce the differing results I got earlier. 
Other plys may be producing nonsense, but at least it's consistent nonsense.
    I see these inconsistencies as a huge problem.

    GNU Backgammon  Position ID: gwEAQCoAAAAAAA
                     Match ID   : cAmgACAAGAAA
     +13-14-15-16-17-18------19-20-21-22-23-24-+     O: wwwcars
     |                  |   | O              O | OOO 2 points
     |                  |   | O              O | OO  
     |                  |   |                  | OO  
     |                  |   |                  | OO  
     |                  |   |                  | OO 
    v|                  |BAR|                  |     5 point match (Cube: 1)
     |                  |   |                  | XX 
     |                  |   |                  | XX  
     |                  |   |                  | XX  
     |                  |   |                  | XX  On roll
     |                  |   | X  X  X     X    | XXX 3 points
     +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: Anterico
      

    Cube analysis
    3-ply cubeless equity  +0.386 (Money:  +0.386)
      0.693 0.000 0.000 - 0.307 0.000 0.000
    Cubeful equities:
    1. No double            +0.599
    2. Double, pass         +1.000  ( +0.401)
    3. Double, take         +0.541  ( -0.059)
    Proper cube action: No double, take (12.8%)
     
    Rollout details:
    Centered 1-cube:
      0.728 0.000 0.000 - 0.272 0.000 0.000 CL  +0.455 CF  +0.647
     [0.000 0.000 0.000 - 0.000 0.000 0.000 CL   0.001 CF   0.001]
    Player wwwcars owns 2-cube:
      0.732 0.000 0.000 - 0.268 0.000 0.000 CL  +1.529 CF  +0.847
     [0.000 0.000 0.000 - 0.000 0.000 0.000 CL   0.000 CF   0.001]
    Full cubeful rollout with var.redn.
    20736 games, Mersenne Twister dice gen. with seed 867000880 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]
     
     GNU Backgammon  Position ID: gwEAQCoAAAAAAA
                     Match ID   : cAmgACAAGAAA
     +13-14-15-16-17-18------19-20-21-22-23-24-+     O: gnubg
     |                  |   | O              O | OOO 2 points
     |                  |   | O              O | OO  
     |                  |   |                  | OO  
     |                  |   |                  | OO  
     |                  |   |                  | OO 
    v|                  |BAR|                  |     5 point match (Cube: 1)
     |                  |   |                  | XX 
     |                  |   |                  | XX  
     |                  |   |                  | XX  
     |                  |   |                  | XX  On roll
     |                  |   | X  X  X     X    | XXX 3 points
     +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: Neil
     
    Cube analysis
    3-ply cubeless equity  +0.420 (Money:  +0.420)
      0.710 0.000 0.000 - 0.290 0.000 0.000
    Cubeful equities:
    1. Double, take         +0.677
    2. Double, pass         +1.000  ( +0.323)
    3. No double            +0.599  ( -0.077)
    Proper cube action: Double, take

    Rollout details:
    Centered 1-cube:
      0.728 0.000 0.000 - 0.272 0.000 0.000 CL  +0.455 CF  +0.647
     [0.000 0.000 0.000 - 0.000 0.000 0.000 CL   0.001 CF   0.001]
    Player gnubg owns 2-cube:
      0.735 0.000 0.000 - 0.265 0.000 0.000 CL  +1.549 CF  +0.877
     [0.000 0.000 0.000 - 0.000 0.000 0.000 CL   0.000 CF   0.000]
    Full cubeful rollout with var.redn.
    20736 games, Mersenne Twister dice gen. with seed 867000880 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]


      ----- Original Message ----- 
       
      From:  Michael  Petch <mailto:[email protected]>  
       
      To: Neil Robins <mailto:[email protected]>  
       
      Cc: [email protected] 
       
      Sent: Thursday, August 27, 2009 9:54  PM
       
      Subject: Re: Requested gnubgautorc
       


      Interestingly enough, after doing some experiments  (with evaluations, 
not rollouts), I am unsure there is a "bug" here. The  position seems very 
volatile if you factor in potential future cube decisions.  It may just be a 
fluke that 1 and 2 ply got it right (It didn't look deep  enough to see future 
cube actions). I'd like to hear feedback from others on  this.

      With that in mind I asked GnuBG to tell me (as an experiment)  what the 
result would be if the checker play evaluations on cube decisions  were 
cubeless, and except for 0 ply everything came out Double/take.

      On  a side note, and I am wondering how other people feel about this. On 
the Hint  screen (or on analysis pane for analyzed positions) you have the 
"0/1/2/3/4"  buttons Each corresponds to a "cubeful N play evaluation". I have 
always found  this not be be very intuitive. My expectation would be that 
"0/1/2/3/4" would  use the existing settings that you can see with "..." and 
simply change the  ply level for cube and checker (and keep all other settings 
like  cubeful/cubeless/noise/filter the same). I found myself wanting to do 
cubeless  checker play evaluations on cube decisions and I kept having to click 
"..."  change the ply level manually, click OK, then hit the Eval button. My 
view is  that if I hit those buttons "0/1/2/3/4" I want to see the difference 
at each  ply level with respect to  my current eval  settings.

      Michael

      On 27/08/09 2:24 PM, "Neil Robins" <[email protected]>  wrote:

       

        I have same  result as initially on a different computer with 20090612  
version.



          ----- Original Message -----  
           
          From:  Michael  Petch <mailto:[email protected]>   
           
          To: Neil Robins <mailto:[email protected]>   
           
          Sent: Thursday, August 27, 2009 9:05   PM
           
          Subject: Re: Requested  gnubgautorc
           


          Something is very bizarre. I'm going to  try some  experiments with 
some internal features of Gnubg turned  off. But as you  suggested in your 
post, exiting Gnubg and restarting  alters the outcomes. but  then I have found 
sometimes all plies start  giving the right output.

          On  27/08/09 1:43 PM, "Neil Robins"  <[email protected]>   
wrote:

           
           

            Strangely, I  am  yet to see a problem with any other position.





_______________________________________________
Bug-gnubg mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/bug-gnubg

Reply via email to