about AEGT

[Uri Blass]

Geschrieben von:/Posted by: Uri Blass at 30 September 2004 16:35:58:

I was asked to tell which version of movei to use in AEGT after I sent 2 versions.
Unfortunately I do not know which version to use and I am not ready for AEGT
so I leave the testers to decide which version to use.
Both 00_8_263 and 00_8_281 scored near 50% against 00_8_251s in my tests
00_8_263 scored slightly more than 50% but the result was not significant.
00_8_281 seem to get 21.5/44 against 00_8_251s at 10 minutes/40 moves at this moment.
I leave the testers to decide which version to use.

There are a lot of options to change the personality
both futilityprune 200 800 1600 2600 4000 and
futilityprune 300 800 1800 3200 5000 performed better than the default against gothmog in my tests.
It suggest that 300 800 1600 2600 4000 may be even better.
The default is 200 800 1800 3200 5000.
Again I am not sure if they are better than the default and there were not enough games to be sure of nothing.
Note that I did not test these parameters with 00_8_281 but only with 00_8_263 and except possible statistical noise it is possible that results are different with 00_8_281 and I tested nothing with 00_8_281 except the default.
Uri

[Heinz van Kempen]

Geschrieben von:/Posted by: Heinz van Kempen at 30 September 2004 16:50:53:
Als Antwort auf:/In reply to: about AEGT geschrieben von:/posted by: Uri Blass at 30 September 2004 16:35:58:

I was asked to tell which version of movei to use in AEGT after I sent 2 versions.
Unfortunately I do not know which version to use and I am not ready for AEGT
so I leave the testers to decide which version to use.
Both 00_8_263 and 00_8_281 scored near 50% against 00_8_251s in my tests
00_8_263 scored slightly more than 50% but the result was not significant.
00_8_281 seem to get 21.5/44 against 00_8_251s at 10 minutes/40 moves at this moment.
I leave the testers to decide which version to use.

There are a lot of options to change the personality
both futilityprune 200 800 1600 2600 4000 and
futilityprune 300 800 1800 3200 5000 performed better than the default against gothmog in my tests.
It suggest that 300 800 1600 2600 4000 may be even better.
The default is 200 800 1800 3200 5000.
Again I am not sure if they are better than the default and there were not enough games to be sure of nothing.
Note that I did not test these parameters with 00_8_281 but only with 00_8_263 and except possible statistical noise it is possible that results are different with 00_8_281 and I tested nothing with 00_8_281 except the default.
Uri

Hello Uri,
okay, I think we will use then Movei 00_8_263. A decision finally has to be taken, as the tournaments are starting.

You can never be sure of anything. Statistics are weird all the time and sometimes fool us. AEGT with additional gauntlets gives more games for amateur engines than any other tournament with long timecontrol I am aware of.
A method to have a better rating estimation for a version would be to let it participate in three AEGT events and only update twice a year. Another method might be to combine results from all important tournaments.
More information concerning AEGT 2 starting tomorrow and again giving 132 games 40/40 and more to all particiants, about the Knight Class Promotion Tournament and the new classes formed according to rating (what was voted unanimously) can be found in about two hours in the new Winboard Forum.
http://www.husvankempen.de/nunn/
Best Regards
Heinz