shrapnel wrote:I feel no excitement.....no real motivation for buying it as the free Stockfish is at least as strong if not stronger than Komodo 7.
Oh, I'll get hold of it later, but not going to be jumping thru hoops to get it quickly.
This is what the "free' Stockfish does to commercial Engines like Komodo and Houdini.
Soon, there will be no motivation left for the commercial Engine developers and fresh young talent graduating from Engineering Colleges will avoid getting into this line.
Congratz Stockfish Development Team !
Yes communists took over. But it is a good thing chess programs are so expensive (not).
I do not think that the developers of stockfish are communists and I expect to see in the future something that is stronger than stockfish.
First things that developers should know in order to get success is to understand what are the main factors for the strength of stockfish.
It can be done by the following way:
take stockfish and start testing simplifications that do not reduce too much elo.
The rules can be that it is better to remove 10 elo for a simplification that reduce 100 lines of code and use SPRT to make sure that you do not remove more elo per line of code(maybe we can even be more aggresive in reduction and decide that we accept reduction of less than 1 elo per line but I think that it is better first to start with allowing only small reductions in strength per line).
Hopefully if you do it you will get something that is more than twice smaller than stockfish and not more than 200 elo weaker when every line is important based on testing because all the tries to remove a line without making it clearly weaker fail.
It can help us to know the important code of stockfish that makes it really strong.
What is your opinion (or anyone else on SF team) as to what makes it strong, other than things in common with the Ippos? I can't figure it out; it seems to me that SF does many things wrong, yet it is still super strong.
An expert in ippos and sf is wanted. Komodo probably make many things right yet still weaker than sf . What are some of the ideas in Komodo that is not in sf?
Exactly right; I'm sure we do many things better than SF, but it seems that there must be things that SF does better than Komodo. Usually when we try to put SF ideas into Komodo (or for fun, Komodo ideas into SF) they do not help, but I'm sure that if we tried all of our original ideas in SF some would help and SF would gain at least 20 elo. But I can't tell you our secrets, I can just say that regarding eval, we pay attention to having things make "chessic" sense as well as test results.
A lingering problem in Komodo seems to be its 'sub par' king safety. It is too slow to see deep threats that result from flank attacks. This makes it harder to trust its vaunted evaluation.
You can argue that it does perform better (relatively and absolutely) on TCEC-like hardware, but the average customer does not have such high end gear, (as cost can be prohibitive, not to mention other factors, like electricity consumption, for example.)
Does Komodo 7 boast any improvements in the area of king safety?
You misunderstand. We have already been doing that for a long time; I just tested to see that if we take it out we only lose 16 elo while SF loses 52 elo for doing the same. We still have to learn why this is the case.
lkaufman wrote:
Does Komodo 7 boast any improvements in the area of king safety?
Regards,
CL
Yes, but nothing of a dramatic nature. As always the problem is that when we "improve" king safety/king attacks, elo usually goes down.
OK, Larry, but may I suggest adding a King Safety parameter then, so the end users (customers) who are interested in a more well-rounded commercial engine can tune king safety to their satisfaction, at least?
With Stockfish being free, you will need to have such features that appeal to the potential customer. Strength is not the only thing potential buyers are interested in. Either way, what I'm suggesting should not impact Komodo's Elo as far as tests go, since testers will use default settings for that.
Komodo is pretty good in unbalanced positions such as queen and knight against two rooks and knight etc. Anyway, there is serious weakness. Komodo is tactically completly blind. That is the reason why is Stockfish much better engine.
My personal experience is that Komodo is useless for analysis (e.g. Freestyle games). I remember just one game where we got some benefit from Komodo usage. Golden standard for analysis is Stockfish + Houdini. Sometimes it is worth to use Gull.
lkaufman wrote:Drawscore was set to zero since our goal is just to win the match, not to rack up a big score. The default -7 is recommended for playing engines other than Stockfish and Houdini 4.
Table size is increased to 256 (from 64) because we think that is best with long time controls, many cores, and big hash tables. Perhaps 512 would be even better, no practical way to know.
My experience with Komodo at the playchess server:
Drawscore ZERO is no good idea. I had clearly more drawish and weaker results than with the default drawscore -7.
And 90% of the games I play and played there was vs Stockfish versions.
lkaufman wrote:Drawscore was set to zero since our goal is just to win the match, not to rack up a big score. The default -7 is recommended for playing engines other than Stockfish and Houdini 4.
Table size is increased to 256 (from 64) because we think that is best with long time controls, many cores, and big hash tables. Perhaps 512 would be even better, no practical way to know.
My experience with Komodo at the playchess server:
Drawscore ZERO is no good idea. I had clearly more drawish and weaker results than with the default drawscore -7.
And 90% of the games I play and played there was vs Stockfish versions.
regards, Clemens Keck
Do you have less than 50% with draw score 0?
draw score 0 is supposed to help relative to default only if you have less than 50%
lkaufman wrote:
Does Komodo 7 boast any improvements in the area of king safety?
Regards,
CL
Yes, but nothing of a dramatic nature. As always the problem is that when we "improve" king safety/king attacks, elo usually goes down.
OK, Larry, but may I suggest adding a King Safety parameter then, so the end users (customers) who are interested in a more well-rounded commercial engine can tune king safety to their satisfaction, at least?
With Stockfish being free, you will need to have such features that appeal to the potential customer. Strength is not the only thing potential buyers are interested in. Either way, what I'm suggesting should not impact Komodo's Elo as far as tests go, since testers will use default settings for that.
Thanks,
CL
We already have a "ktabm" opening parameter in Komodo tcec, which will remain in Komodo 7, that affects king safety, although we could add other parameters. But of course it's not just a matter of getting the parameter values right, but also of the definition of king safety.
. What are some of the ideas in Komodo that is not in sf?
