Geschrieben von: / Posted by: Uri Blass at 23 February 2004 22:43:30:
Als Antwort auf: / In reply to: Re: PostModernist v1010 Now Available geschrieben von: / posted by: Andrew Williams at 23 February 2004 21:29:50:
You can download this latest version from:
http://homepage.ntlworld.com/postmodernist/The relevant section from the README file is presented below. Roughly translated, it says that PMv1009 was an Abomination From The Swamp. If your tourney's rules allow it, update now!
1. Changes
==========
v1010 - Fixed another bug in the evaluation (iqp_white). This
caused mismatches between white and black scores for
a given position. This one was pretty serious.
- Removed the recapture extension change from v1009. This
was removed while trying to isolate the evaluation bug.
It will be tested again for version 1011.
Andrew
Hello Andrew,
Thanks for your upgrade.
Good new for you , PostModernist v1009 has defeated The Butcher 1.42c in my tournament this afternoon and PostModernist has 3/3 points (won vs Go2Chess 0.12 and Dragon 4.5 CF)
The 4th round will be played by PMv1010 or PMv1011
Best wishes
Claude
Thanks for this good news! In testing, PMv1009 did very well in blitz games, but significantly worse in longer games. It's probably not THAT bad, but it does lose heavily to v1007 whereas v1010 does very much better.
Cheers
Andrew
I also discovered right now that I had a significant bug in many versions of movei including the versions from CCT6 and Movei00_8_158
It turned out that movei used swap_list[-1] in the function to calculate see.
It did not prevent it to play a lot of games with almost no problem and I saw no stupid blunders and better results in blitz based on my tests against itself.
It scored slightly worse in my tests at longer time control against Junior but I thought that it may be a statistical error(espacially when I do not have a lot of computer time to test and it also scored very well in CCT6 inspite of the bug).
I was too lazy to use checking bounds or even to run it on debug mode to compare number of nodes with release mode and only now after I found different number of nodes after a change that was supposed to change nothing except small changes in speed I discovered the problem.
Uri