50 Elo from special endgame heuristics?

[BB+]

http://www.stmintz.com/ccc/index.php?id=473034

Rybka 1.0 doesn't have any special endgame heuristics. This can be easily fixed. If I had to make an ELO guess, I'd go with something like 50.

Anyone else think it could be this high? I can't see that he's done too much work on this in the last 6 years.

[hyatt]

Hard to say. There are two kinds of endgame stuff. The kind that handles things like outside passed pawns and stuff and which says something like "this will be good with less material on the board, to head for the endgame". The other kind is actual endgame code. In Crafty, I added KRP KR code to avoid relying on egtbs which I don't particularly like due to the performance issues. Recognizing that some endgame features should make you head to the endgame might get up to 50 although that seems high. Some endgame evaluation terms can be quite valuable. Passed pawn advances come to mind. I suppose it is possible that the whole kit and kaboodle could add 50 or more elo to a program that started out with nothing.


I suppose it is possible to go thru my eval and disable anything endgame related and do a cluster test. Will try to do that this week after classes are started...