buggnubg - Bugs: bug #25409, Stop when STDs are small enough...

 
 

bug #25409: Stop when STDs are small enough doesn't work

Submitted by:  Philippe Michel <plm>
Submitted on:  Sun 25 Jan 2009 10:51:28 PM UTC  
 
Originator Name: Philippe MichelItem Group: Other bug
Category: RolloutsRelease: 0.90-devel
Privacy: PublicOperating System: Linux
OS Type: NoneProcessor Type: None
Severity: 3 - NormalOpen/Closed: Closed
Assigned to: NonePriority: 5 - Normal
Status: Fixed

Add a New Comment(Rich Markup)
   

You are not logged in

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

 

Tue 27 Jan 2009 12:16:53 PM UTC, comment #1:

Should be fixed now. Notice that the limit refers to all stats (win/g/bg...cl/cf). This functionality may well be broken in the non mt code as well. I've recently fixed the jsd for cube rollouts, so you may want to use that instead.

Christian Anthon <c_anthon>
Project Administrator
Sun 25 Jan 2009 10:51:28 PM UTC, original submission:

I try to run rollouts with gnubg -t -c with this kind of command file :

set threads 1
set matchid cAkAAAAAAAAA
set board 3gMAAG27OwAAAA
set rollout trials 46656
set output digits 5
set rollout truncation enable off
set rollout jsd stop off
set rollout jsd move off
set rollout limit enable on
set rollout limit minimumgames 1296
set rollout limit maxerr 0.0040
hint
rollout =cube
export position text ...

But the "rollout limit" parameters don't seem to be effective. Usually the it doesn't stop when the stds are small enough, for instance :

46656 games will be played per rollout.
Probabilities and equities will be shown with 5 digits after the decimal separator
Games in rollouts will be played out until the end.
Do not stop rollout based on JSDs
Do not stop rollout of moves based on JSDs
Stop rollout when STD's are small enough
After 1296 games, rollouts will stop if the STDs are small enough.
Rollouts can stop when the ratio |STD/value| is less than 0.0040 for every value (win/gammon/backgammon/...equity)

Cube analysis
2-ply cubeless equity +0.50483
0.75242 0.00000 0.00000 - 0.24758 0.00000 0.00000
Cubeful equities:
1. Double, take +0.84191
2. Double, pass +1.00000 ( +0.15809)
3. No double +0.83320 ( -0.00872)
Proper cube action: Double, take

Rollout: 4663/46656 trials
Centered 1-cube:
0.76088 0.00000 0.00000 - 0.23912 0.00000 0.00000 CL +0.52176 CF +0.88065
[0.00027 0.00000 0.00000 - 0.00027 0.00000 0.00000 CL 0.00054 CF 0.00155] r
Time elapsed 18m46s Estimated time left 2h48m39s
Estimated SE for "Centered 1-cube" after 46656 trials 0.00049
Player gnubg owns 2-cube:
0.76103 0.00000 0.00000 - 0.23897 0.00000 0.00000 CL +0.52205 CF +0.89274
[0.00027 0.00000 0.00000 - 0.00027 0.00000 0.00000 CL 0.00054 CF 0.00156] r

All stds are well under 0.004. The rollout should have stopped somewhere between 1296 and there.

On the other hand, sometimes it stops at 1 game :

Rollout details:
Centered 1-cube:
0.73834 0.00000 0.00000 - 0.26166 0.00000 0.00000 CL +0.47667 CF +0.59439
[0.00000 0.00000 0.00000 - 0.00000 0.00000 0.00000 CL 0.00000 CF 0.00000]
Player gnubg owns 2-cube:
0.73836 0.00000 0.00000 - 0.26164 0.00000 0.00000 CL +0.95345 CF +0.60303
[0.00000 0.00000 0.00000 - 0.00000 0.00000 0.00000 CL 0.00000 CF 0.00000]
Full cubeful rollout with var.redn.
1 games, Mersenne Twister dice gen. with seed 810709162 and quasi-random dice
Stop when std.errs. are small enough: ratio 0.004 (min. 1296 games)

This is the same executable than the first example.

Philippe Michel <plm>
Project Member

 

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

Attach File(s):
   
   
Comment:
   

No files currently attached

 

Depends on the following items: None found

Items that depend on this one: None found

 

Carbon-Copy List
  • -unavailable- added by c_anthon (Posted a comment)
  • -unavailable- added by plm (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 2 latest changes.

    Date Changed By Updated Field Previous Value => Replaced By
    Tue 27 Jan 2009 12:16:53 PM UTCc_anthonOpen/ClosedOpen=>Closed
      StatusNone=>Fixed

    Back to the top


    Powered by Savane 3.1-cleanup1