OpenChess

Is Houdini's eval more simple and less knowledgeable than IvanHoe's eval, thus trading knowledge for speed in the search function? Are there any easy ways, test suites, etc., to determine these things about an engine?

My tests show that Houdini is clearly better than Rybka 4 at short time controls -- ( and i was very skeptical of its dominance ) .
I think there are some preliminary tests that suggest Rybka 4 dominance at longer time controls.

The rybka fans have the theory that Rybka 4 and Rybka 3 have a lot more chess knowledge than any ippolit engine


I suggested also that Rybka will beat Houdini/ippolit at long time control that ippolit cannot be improved by much.

Long time control tests may bring out if RYbka 4 has more chess knowledge.

So an even more interesting question is : is Ippolit/Houdini dumber than Rybka !!!

benstoker wrote:Is Houdini's eval more simple and less knowledgeable than IvanHoe's eval, thus trading knowledge for speed in the search function? Are there any easy ways, test suites, etc., to determine these things about an engine?

Houdini has a more sophisticated evaluation function (with more terms and more knowledge) than Ivanhoe, as well for the middle game as for the end game.

Robert

The rybka fans have the theory that Rybka 4 and Rybka 3 have a lot more chess knowledge than any ippolit engine ...
So an even more interesting question is : is Ippolit/Houdini dumber than Rybka !!!

There is more "knowledge" in R3 than IPPOLIT, but a lot of that (from my correspondence with LK) is more aimed for anti-human play. Also, as VR himself said, it's not really "knowledge" if it doesn't make the engine play better. There are a nominal amount of additional positional features that R3 considers which do not appear in IPPOLIT (and a few vice-versa, not to mention the "obvious" things such as KPK scoring, BN mate, blind bishop, bishop underpromotion...). The situation with R4 is different, as the evaluation function has notably changed (the accounting is not easy, but I think it is smaller overall).

Charles wrote: . . . So an even more interesting question is : is Ippolit/Houdini dumber than Rybka !!!

The sentence "Is Jessica Biel more sexy looking than Zach Galifianakis?" is logically equivalent to "Is Zach Galifianakis more sexy looking than Jessica Biel?", right? The truth value of one proposition is merely the negation of the other.

So an even more interesting question is: How is mere negation more interesting? Unless of course you're a philosophy professor of symbolic logic who teaches logic, sets and functions courses to weed out computer science majors.

Robert Houdart wrote:

benstoker wrote:Is Houdini's eval more simple and less knowledgeable than IvanHoe's eval, thus trading knowledge for speed in the search function? Are there any easy ways, test suites, etc., to determine these things about an engine?

Houdini has a more sophisticated evaluation function (with more terms and more knowledge) than Ivanhoe, as well for the middle game as for the end game.

Robert

Hello Robert,
could you tell us which areas of IvanHoe's eval you improved more?
Thanks for Houdini!