Leto wrote:Thanks for running these matches. It seems that Houdini 3 is still king in all time controls.
Houdini 3 has indeed come out ahead in these matches, but the margins have been so close (at all but blitz levels) that I would be willing to wager a thousand dollars that some version of Komodo will defeat Houdini 3 in similar matches by the end of this year, assuming that Clemens (or another trusted person) is willing to run such matches and hold the stakes.
Alas for Komodo I think there will be a Houdini 4 by then...
For Captain Larry I'll make a special "Moby Dick" compile of Houdini .
Captain Larry: From hell's heart I stab at thee; for hate's sake I spit my last breath at thee. Ye damned whale.
Leto wrote:Thanks for running these matches. It seems that Houdini 3 is still king in all time controls.
Houdini 3 has indeed come out ahead in these matches, but the margins have been so close (at all but blitz levels) that I would be willing to wager a thousand dollars that some version of Komodo will defeat Houdini 3 in similar matches by the end of this year, assuming that Clemens (or another trusted person) is willing to run such matches and hold the stakes.
Alas for Komodo I think there will be a Houdini 4 by then...
For Captain Larry I'll make a special "Moby Dick" compile of Houdini .
Leto wrote:Thanks for running these matches. It seems that Houdini 3 is still king in all time controls.
Houdini 3 has indeed come out ahead in these matches, but the margins have been so close (at all but blitz levels) that I would be willing to wager a thousand dollars that some version of Komodo will defeat Houdini 3 in similar matches by the end of this year, assuming that Clemens (or another trusted person) is willing to run such matches and hold the stakes.
Alas for Komodo I think there will be a Houdini 4 by then...
For Captain Larry I'll make a special "Moby Dick" compile of Houdini .
I do have the compulsion to see Komodo surpass Houdini, so it's a good analogy in that respect. But if my memory is correct regarding a book I read over half a century ago, I think the motivation was hatred. I certainly don't hate Robert (actually I can't think of anyone I really hate other than obvious choices like murderous tyrants). I'll never think highly of his integrity due to claiming the original Houdini to be his original engine, when it was just a copy of some Ippo version with only trivial changes, but I think highly of his skills as a programmer and consider him a worthy rival for Don and myself. Even if Stockfish or some other engine with no ethical issues were the best, we would be trying hard to be number one.
M ANSARI wrote:Again! Houdini 3 really needs some more heuristics when it comes to rook endgames. It plays way below its level in rook endgames. I guess against weaker opponents or at faster time controls being down a pawn in a rook endgame is no problem, but at LTC with stronger opponents it can be very different. This is really becoming a weakness with Houdini 3 and my guess is that the engine is much stronger if this little backdoor can be fixed.
I agree, but it does take another strong program to expose this weakness. Robert may not have been able to detect this weakness using only ultra fast test games during H3 development.
Actually Houdini was losing a lot of games against Rybka 4.1 in a similar manner. It just seems to lose too many games to this one weakness and that probably makes it lose a lot of cheap ELO points that can be avoided with that one fix. But of course I could be wrong and the rook endgame evaluations could be scoring well during testing via hundreds of thousands of games and changing that could have a negative impact. However if you look at H3 games that it loses ... just way too many games are lost cheaply to that one weakness. It is losing games that it should simply not lose and that an engine rated 400 ELO points lower would easily be able to hold. Houdini initially had a weakness by over evaluating the value of the queen, but that was fixed. My bet is the Robert is also seeing this and might even have fixed it in H4 beta.
.
