buggnubg - Bugs: bug #33917, "No double" evaluation...

 
 

bug #33917: "No double" evaluation exceeds 1.000 even though opponent has 11 men off.

Submitted by:  Tom Martin <tom_martin>
Submitted on:  Mon 01 Aug 2011 01:05:42 PM UTC  
 
Originator Name: Tom MartinItem Group: Other bug
Category: EvaluationsRelease: 0.90-devel
Privacy: PublicOperating System: All
OS Type: AnyProcessor Type: Any
Severity: 3 - NormalOpen/Closed: Closed
Assigned to: NonePriority: 5 - Normal
Status: Postponed

Add a New Comment(Rich Markup)
   

You are not logged in

Please log in, so followups can be emailed to you.

 

Wed 03 Aug 2011 09:40:55 PM UTC, comment #1:

I don't think this is related to Janowski's calculations. The equity higher than 1 in the no-double branch comes from bad takes later in the rollout and these most likely come from inaccurate evaluations by the neural net.

Using 2-ply cube for the rollout may help to avoid some of them.

Regarding your last sentence, it is likely the double branch underestimates the doubler's equity since he faces difficult containment or trapping plays while his opponent's are often forced. Probably using 2-ply checker play, ideally with a large move filter, would give a higher equity, maybe a pass.

The no-double equity is obviously wrong, and gnubg could detect that, but then what ? There is no easy way to estimate the real equity.

I'm quite confident this is not an "oops" bug in cubeful calculations but a weakness of the neural net and there is no easy fix for that.

Philippe Michel <plm>
Project Member
Mon 01 Aug 2011 01:05:42 PM UTC, original submission:

There seems like a bug in the application of Janowski's cubeful equity algorithm. It should never exceed 1 if no gammon is possible, right? If you let it show a value greater than 1 for some sort of educational reason, it shouldn't affect the cube decision. The attached file shows ND/T despite an estimated equity of .970 after the take. If both evaluations are uniformly too high, then ND/T makes sense, but it's definitely confusing.

Tom Martin <tom_martin>

 

(Note: upload size limit is set to 16384 kB, after insertion of the required escape characters.)

Attach File(s):
   
   
Comment:
   

Attached Files
file #23724:  29-24$.sgf added by tom_martin (2KiB - application/octet-stream)

 

Depends on the following items: None found

Items that depend on this one: None found

 

Carbon-Copy List
  • -unavailable- added by plm (Posted a comment)
  • -unavailable- added by tom_martin (Submitted the item)
  •  

    Do you think this task is very important?
    If so, you can click here to add your encouragement to it.
    This task has 0 encouragements so far.

    Only logged-in users can vote.

     

    Please enter the title of George Orwell's famous dystopian book (it's a date):

     

     

    Follow 5 latest changes.

    Date Changed By Updated Field Previous Value => Replaced By
    Thu 04 Aug 2011 10:04:34 PM UTCplmOpen/ClosedOpen=>Closed
      StatusNone=>Postponed
    Wed 03 Aug 2011 09:40:55 PM UTCplmOperating SystemWindows XP=>All
      OS Type64Bit=>Any
    Mon 01 Aug 2011 01:05:42 PM UTCtom_martinAttached File-=>Added 29-24$.sgf, #23724

    Back to the top


    Powered by Savane 3.1-cleanup1