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

BGonline.org Forums

xg vs gnubg containment position

Posted By: Ian Dunstan
Date: Friday, 1 August 2014, at 2:22 p.m.

In Response To: xg vs gnubg containment position (Bob Koca)

Do you mean in general or just positions similar to the one in your post?

All my rollouts were similar to the one I posted in that they were cubeless with the same settings. I had no idea "cubeless" rollouts, as I was doing them for XG, meant that the bot was playing it's moves like it thought it should be able to turn the cube, but never actually be allowed to turn it. That is what I make of eXtreme's statement posted earlier and shown below:

There is a difference between GnuBG cubeless RO and XG's.

In GnuBG it basically assume the cube cannot be turn and will play checker play accordingly.

In XG, it knows the cube can be used and will play the checker accordingly, but it will never double

I have just redone a rollout of a backgame position which originally took me 5184 trials to accuracy match gnubg's 1296. Now the two bots approximately agree with each other (CI wise) after the same number of trials. It seems from what you report, and my latest XG rollout, that doing the cubeless rollouts the way I was doing them confuses poor XG quite badly.

Oh, well. It looks like lesson 1 in learning to use XG effectively has been finished for me. I'm glad XG's variance reduction now appears similar to gnubg's, I was disappointed to think otherwise before.

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.