Andscacs - New version 0.71

[cdani]

Hi!
I published a new version of Andscacs, 071.

Changes:

* Improved evaluation of passed pawns.
* General improvement of speed.
* Fixed some bugs.
* New king safety evaluation.
* Little improvements on LMR.
* Improved mobility values.
* Removed second hash of evaluations. Now uses only main hash for this.

It seems something more than 40-50 elo stronger. I think will be like this also in slow time controls; seems it is.

It's getting more and more difficult to improve it with more or less easy to find stuff. May be I will need to do more fine grained work .

http://www.andscacs.com/

[Graham Banks]

Getting stronger all the time!

[cetormenter]

Thanks for the new release Daniel. I want to add Andscacs to my gauntlet but it seems that it does not like quick time controls. Even with a fairly liberal (by engine standards) time control of 1 minute per game, it crashes every 10-20 games. I went back to test 0.70 and the same thing occurs.

[cdani]

Ok. I will try to fix this.
I always use increment, so I did not notice this bug.
Thanks!

[cetormenter]

It occurs even with increment. I tried 6"+0.05" and 12" + 0.05" and it crashes with both. So it seems to just be an issue with low time left on the clock.

[cdani]

I'm not able to reproduce the problem. I'm doing some hundred games with an I7-920 at 10 seconds/game and in an AMD-FX 8350 with 15 seconds/game and no single time loss or crash.

Which cpu are you trying? Are you using cutechess? With how many simultaneous games? May be you can send me some logs?
Thanks!

[cetormenter]

My cpu is an i5-3210M. I am using cutechess 0.6.0 and running two games at the same time. I also just ran similar tests under Arena at 6" + 0.05" per game. The same thing happens. Every 10-20 games it crashes. Perhaps it does not like being run as a UCI engine?

Here is a log file from the Arena run. It seems that Andscacs gets stuck calculating for longer than it should and then it crashes. Search for "on time" to find the most relevant part of the file.


http://www.mediafire.com/download/ac67b ... ile001.log

[cdani]

Aha. I'm using cuteches 0.5.1, that is known to be more stable than 0.6. May be its hiding this behavior to me. Also I don't use Arena.
So I will try to make it work with your version of cutechess.
Anyway as the problem it's only for very short time controls, anyone can use this new version for tournaments if they wish, as has been done with previous version 0.70 without problems.
When I succeed to arrange this, I will publish here.
Thanks!!

[cdani]

Really I'm not able to make new version 0.71 crash or lose on time, even with the last version of cutechess.
In 0.71 I switched to QueryPerformanceCounter to have more precision, and seem to work well for me, better than 0.70 for very fast time controls.

May be your behavior is related to some of this?
http://support.microsoft.com/kb/895980/en-us

[cetormenter]

Well if it is only found in 0.71 and not 0.70 then that is not the issue as I am getting crashing with both versions. Plus I am not using AMD nor windows XP. If nobody else is reporting any problems then I would not worry about it. Thanks for looking into though.