GreKo 2018

[WinPooh]

http://greko.su/greko-2018.zip

GreKo 2018
-----------
31-Dec-2018

*** HAPPY NEW YEAR!!! ***

(new) * Added static evaluation hash.
(new) * Added UCI options for eval and pawn hash size.
(chg) * Optimizations in evaluation weights.

Test matches (60+1):

vs. Fruit 2.1: 25.5 - 14.5
vs. Crafty 23.4: 14.5 - 25.5
vs. Delfi 5.4: 25 - 15

Expected playing strength: +20...30 Elo to 2018.07, about 2740 CCRL.

[Kotlov]

С новым годом, с новым Греко! ))

[CMCanavessi]

I don't understand the version numbering.. shouldn't this be 2019 after we had several 2018.X versions?

[Alexander Schmidt]

I don't understand the version numbering.. shouldn't this be 2019 after we had several 2018.X versions?

Maybe you understand it when you are looking at the date of the announcement

Thanks Vladimir, GreKo is still one of my favorites!

[CMCanavessi]

I don't understand the version numbering.. shouldn't this be 2019 after we had several 2018.X versions?

Maybe you understand it when you are looking at the date of the announcement

Thanks Vladimir, GreKo is still one of my favorites!

I know that, but we already had GreKo 2018.02, GreKo 2018.07, GreKo 2018.08 and GreKo 2018.12... it makes little sense to me to have GreKo 2018 _AFTER_ those... sounds like the other ones were some kind of beta or pre-release... Isn't 2018.02 higher than 2018 after all? Common sense says yes.

[tpoppins]

I must agree with Carlos: having a "2018" version follow "2018.08" and "2018.12" is counter-intuitive and confusing.

Good point about development versions, too: some of the many GreKo versions released last year were in effect just that. In particular v2018.12 appears to be of exactly the same strength as v2018.08 -- a dead-even tie after 60 games and exactly the same score (down to identical proportion of wins, losses and draws) vs. GreKo's derivative Igel. Now that this new "2018" version is out that puts the kibosh on 2018.12 testing, and with so few games run so far it's likely to end up lower on the list than v2018.08.

Nobody wins in such a scenario: the testers waste time and electricity on a version not worth testing, the author doesn't get an accurate estimate of the new version's strength due to the low number of games. A simple line or two in the readme to the effect of "a development version, rating lists probably shouldn't bother" would avoid all that. Not every version must come with an Elo increase; new features and bug fixes can be just as important. The computer chess world would be a sad little place indeed if it were all about Elo.

Happy New Year and may GreKo see many more!