BGonline.org Forums

XG version 2 and the Depreli positions

Posted By: Frank Berger
Date: Sunday, 17 July 2011, at 9:25 a.m.

In Response To: XG version 2 and the Depreli positions (Timothy Chow)

I'm working on an XML file that contains all positions from the Depreli 2010 and told that already in PM to Xavier and Michael Petch. That way you just need a little glue code to run your AI through the whole benchmark.

I put quite some effort in finding inconsistencies and cleaning them up together with Michael Depreli (you have just to try it for your self for some 100 positions, to see how error prone Michaels job has been and how much work it has been. Kudos to Michael) and adding more rollouts. One reason that BGBlitz updates were so few in the last year. Just imagine how long it takes to handle one position and then multiply it with 5000 ....

I regard this as very valuable tool to get a quick evaluation of the playing strength (BGB needs 30 minutes on a moderate fast PC). Due to the large number of positions it is much more valuable than the 2005 benchmark, it will be published, the numbers can be checked, it makes the AI comparable and by avoiding any manual activities it will be very reliable. Further it can be extended with new positions and the rollout values can be improved (there are a couple of positions were I believe the numbers are wrong, mostly deep backgames) leeding to new versions. IMHO something was overdue.

ciao Frank

Messages In This Thread

• XG version 2 and the Depreli positions
  Timothy Chow -- Saturday, 16 July 2011, at 11:36 p.m.
  • XG version 2 and the Depreli positions
    Frank Berger -- Sunday, 17 July 2011, at 9:25 a.m.
    • +100 *NM*
      Michael Petch -- Sunday, 17 July 2011, at 4:40 p.m.