[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

GnuBg settings to mimic XGR++

Posted By: Taper_Mike
Date: Tuesday, 31 July 2012, at 12:36 a.m.

In Response To: Stick vs Merriman: XG++ Data Makes No Sense (Daniel Murphy)

–

Nice work by Daniel.

I think it is a great idea to define (and save) a GnuBg rollout setting that mimics XGR++.

Be careful if you try this. It is easy to screw up the default settings, and if you have not created some sort of go-back plan, you may be stuck with your new settings! I saved the current rollout settings before beginning, but that is not enough. Before messing with the move filter, you need to write down its default settings, and then make sure that any changes you make are only to a “user-defined” move filter.

As a first effort, I tried to mimic Daniel’s rollout in the preceding message. As the rollout below shows, I was not entirely successful. Even though I am using the same seed as Daniel, the equity of the best play is off by 0.01 compared to his rollout.

A couple of things occur to me:

  • Daniel may be using a different version of GnuBg that I am. My version number is “Version GNU Backgammon 0.90-mingw 20120408.”
  • Daniel has different settings for the move filters used by Player 0 and Player 1. These are the filters applied after the first two decisions. Mimicing XGR++ would seem to require using the same move filters for both players. Daniel probably uses his settings for some other purpose, and simply forgot to reset them.
  • I tried to duplicate his settings, but the best I could do is get Player 0 to “keep the first 1000 0-ply moves.” Daniel has him keeping “the first 5995833 0-ply moves.”
  • Daniel has the move filter for Player 1 set to supremo settings: “keep the first 0 0-ply moves and up to 16 more moves within equity 0.32.” It seems odd to use these settings when the first two decisions are being made with the normal (smaller) move filter.
  • Daniel is adding 0.015 determistic noise to moves after the first 2. I am not sure why. Is it desirable to add noise in order to mimic XGR++? My understanding is that adding noise is useful only when you wish to degrade the quality of play.

I also have a question about plys. GnuBg has a setting for “Evaluation for later plys.” Under that heading, I put a checkmark for “Enable separate evaluations,” and then I selected “Change eval ply after:” 2. This seems to be the setting that Daniel used. I wonder, though, if it is the right one.

XGR++ makes its “First 2 decisions in 3-ply, then 2-ply checker and 3-ply cube.”

For GnuBg, is the first decision a 0-ply decision? If so, then the second decision for GnuBg occurs at 1-ply, and any decisions after that should use the weaker eval settings. That would mean setting GnuBg to “Change eval ply after:” 1. I tend to think that Daniel has this one right. It’s just that I am not certain.

This effort had me using a few GnuBg dialogs that I had not seen before, so I cannot be sure I have everything right. Daniel, can you help me out? GnuBg experts?

Mike

The score (after 0 games) is: gnubg 0, MusicLover 8 (match to 11 points)

Move number 5: MusicLover to play 53

gnubg174

' '2X '2X4X ' ' '1O ' '

2O2O3O2O1X4O ' ' '1O '3X
MusicLover72
Position ID: zB5wQHDbzUMQAA Match ID: UYluAQAAQAAE

• MusicLover moves 10/5*/2

Cube decision
2-ply cubeless equity +1.325556(Money: +1.76092 +1.325556)
0.940369 0.841438 0.040789 - 0.059631 0.002037 0.000005
Cubeful equities:
1.No double +1.154099
2.Double, take +1.043155 -0.110944
3.Double, pass +1.000000 -0.154099
Proper cube action:Too good to redouble, pass

# Ply Move Equity
1 R 15/10 6/3 +1.279891
0.9438960.8827450.042611-0.0561040.0007830.000005 +1.356112 +1.279891
0.0010830.0016880.001566-0.0010830.0000920.000002 0.002795 0.004160
Truncated cubeful rollout (depth 7) with var.redn.
360 games, Mersenne Twister dice gen. with seed 691526212 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]
Different evaluations after 2 plies:
Player 0:
Play: 1-ply cubeful prune, noise 0.015 (d)
keep the first 1000 0-ply moves and up to 8 more moves within equity 0.16
Cube: 2-ply cubeful prune, noise 0.015 (d)
Player 1:
Play: 1-ply cubeful prune, noise 0.015 (d)
keep the first 0 0-ply moves and up to 16 more moves within equity 0.32
Cube: 2-ply cubeful prune, noise 0.015 (d)
• 2 R 10/5*/2 +1.268968 ( -0.010922)
0.9451590.8882600.068927-0.0548410.0006640.000003 +1.361720 +1.268968
0.0011580.0018930.001856-0.0011580.0001070.000006 0.003049 0.004383
Truncated cubeful rollout (depth 7) with var.redn.
360 games, Mersenne Twister dice gen. with seed 691526212 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]
Different evaluations after 2 plies:
Player 0:
Play: 1-ply cubeful prune, noise 0.015 (d)
keep the first 1000 0-ply moves and up to 8 more moves within equity 0.16
Cube: 2-ply cubeful prune, noise 0.015 (d)
Player 1:
Play: 1-ply cubeful prune, noise 0.015 (d)
keep the first 0 0-ply moves and up to 16 more moves within equity 0.32
Cube: 2-ply cubeful prune, noise 0.015 (d)
3 R 15/7 +1.261523 ( -0.018368)
0.9391380.8831720.044209-0.0608620.0020860.000026 +1.345183 +1.261523
0.0012160.0016410.001858-0.0012160.0003760.000007 0.003111 0.004429
Truncated cubeful rollout (depth 7) with var.redn.
360 games, Mersenne Twister dice gen. with seed 691526212 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]
Different evaluations after 2 plies:
Player 0:
Play: 1-ply cubeful prune, noise 0.015 (d)
keep the first 1000 0-ply moves and up to 8 more moves within equity 0.16
Cube: 2-ply cubeful prune, noise 0.015 (d)
Player 1:
Play: 1-ply cubeful prune, noise 0.015 (d)
keep the first 0 0-ply moves and up to 16 more moves within equity 0.32
Cube: 2-ply cubeful prune, noise 0.015 (d)

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.