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

BGonline.org Forums

Bot evals (XGR bug: no, but surely need to be improved)

Posted By: eXtreme Gammon
Date: Wednesday, 14 December 2011, at 1:32 p.m.

In Response To: Bot evals (XGR bug) (Matt Cohn-Geier)

I've started in the past week to analyze and categorize XG errors using the data generated by the Depreli Study.

The result is that containment position is the largest source of XG errors and also the one where the NN is the less accurate in term of absolute equity (obviously these are tied).

One of the reason XGR++ get that wrong is that containment position are often long games, and the XGR truncation (after 5 moves) happens most of the time when the position is still a containment one.

In the position you show, XGR++ shows its limit as the moves played using 2-ply clearly generate some large errors.

This is the same than what happen in that thread: http://www.bgonline.org/forums/webbbs_config.pl?frames;read=110862

I'll be working on that aspect of the engine in the future.

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.