Page 10 of 11

Re: To Larry Kaufman

Posted: Mon Mar 14, 2016 9:24 am
by Nordlandia
Fritz 15 use alot of more paramaters and still work in majority of GUIs

Image

Re: To Larry Kaufman

Posted: Mon Mar 14, 2016 10:12 pm
by TShackel
Larry,

I know that elo is always the most important when deciding what changes to make to Komodo, but I wondered if you and mark both intentionally try to change/improve the style of komodo when making changes?

If so, how would you characterize any stylistic improvements you may have made based on the games of 9.3 and the games of 9.4? Is Komodo a better attacker with 9.4, or better at closed positions? Is it better at openings or endgames, and how aggressive is it? Is it a better looking player in slow strategical squeezes or in a tactical slugfest, or both?

Those are some of the questions I would have about what stylistic changes you may have made. I'm not asking if you can magically predict komodo's style changes after making a tiny change to its code without seeing any games with the new version, but I ask you because I assume you look at many of komodos games, and get a sense for what its style is, and also looking through komodo's losses may give you ideas for improvements to test.

I was thinking about starting a new thread for komodo's style, but since you were in this thread I thought I'd write it here. You don't have to answer if you have nothing to say/add about K's style. I know devveloping a chess engine is about incremental improvements afterall that improve elo, and it's impossible to predict what style changes that may make to komodo. I just wondered if you compared K 9.4 to 9.3 and looked at games from both to discern stylistic differences.

Sincerely,

Tim.

Re: To Larry Kaufman

Posted: Tue Mar 15, 2016 12:56 am
by lkaufman
TShackel wrote:Larry,

I know that elo is always the most important when deciding what changes to make to Komodo, but I wondered if you and mark both intentionally try to change/improve the style of komodo when making changes?

If so, how would you characterize any stylistic improvements you may have made based on the games of 9.3 and the games of 9.4? Is Komodo a better attacker with 9.4, or better at closed positions? Is it better at openings or endgames, and how aggressive is it? Is it a better looking player in slow strategical squeezes or in a tactical slugfest, or both?

Those are some of the questions I would have about what stylistic changes you may have made. I'm not asking if you can magically predict komodo's style changes after making a tiny change to its code without seeing any games with the new version, but I ask you because I assume you look at many of komodos games, and get a sense for what its style is, and also looking through komodo's losses may give you ideas for improvements to test.

I was thinking about starting a new thread for komodo's style, but since you were in this thread I thought I'd write it here. You don't have to answer if you have nothing to say/add about K's style. I know devveloping a chess engine is about incremental improvements afterall that improve elo, and it's impossible to predict what style changes that may make to komodo. I just wondered if you compared K 9.4 to 9.3 and looked at games from both to discern stylistic differences.

Sincerely,

Tim.
The new parameter "Dynamism" will have far more effect on Komodo's style than any changes we have made to the default version. It will allow you to make Komodo as cautious or as wild as you like. As for the default version, we have made some changes to king safety that hopefully will make Komodo a better attacker and defender. Other changes are more subtle.

Re: To Larry Kaufman

Posted: Tue Mar 15, 2016 9:40 am
by PK
One more possible idea: some time ago I noticed that one of the free versions of Komodo has a boolean option forcing it to play the best book move (so that the book move varies only when the book contains a couple of moves with the exact same score). Inspired by that, in my own engine I implemented a BookFilter option, which allows only book moves whose score exceeds certain percentage of the best book move. At 100%, it behaves exactly like your option, but it gives the user more freedom. For example, if a book contains info like 1.d4 40% 1.e4 40% 1.c4 10% 1.Nf3 10%, BookFilter above 25% prunes 1.c4 and 1.Nf3.

Re: To Larry Kaufman

Posted: Tue Mar 15, 2016 5:30 pm
by TShackel
lkaufman wrote:The new parameter "Dynamism" will have far more effect on Komodo's style than any changes we have made to the default version. It will allow you to make Komodo as cautious or as wild as you like. As for the default version, we have made some changes to king safety that hopefully will make Komodo a better attacker and defender. Other changes are more subtle.
Thanks Larry for your thoughtful answer. So the default version only had king safety changed which may affect komodo's attack/defense ability, and then there were other subtle changes that may or may not make a stylistic improvement.

Would you mind describing the value ranges for dynamism? Like what is the range of values for dynamism, and what is the default value? Since you think this setting will change komodo's style greatly, will the default dynamism setting increase the dynamic style of komodo to some degree?

Thanks.

Sincerely,

Tim.

Re: To Larry Kaufman

Posted: Tue Mar 15, 2016 11:35 pm
by lkaufman
TShackel wrote:
lkaufman wrote:The new parameter "Dynamism" will have far more effect on Komodo's style than any changes we have made to the default version. It will allow you to make Komodo as cautious or as wild as you like. As for the default version, we have made some changes to king safety that hopefully will make Komodo a better attacker and defender. Other changes are more subtle.
Thanks Larry for your thoughtful answer. So the default version only had king safety changed which may affect komodo's attack/defense ability, and then there were other subtle changes that may or may not make a stylistic improvement.

Would you mind describing the value ranges for dynamism? Like what is the range of values for dynamism, and what is the default value? Since you think this setting will change komodo's style greatly, will the default dynamism setting increase the dynamic style of komodo to some degree?

Thanks.

Sincerely,

Tim.


Well, the default is just 100% of what we believe makes for strongest play. The max is 200% (i.e. double), the min is zero, but if you set it to zero it will play silly, and even a setting like 1 or 2 will give ridiculous play. Maybe something like 10 is about the lowest value where it will play somewhat plausibly. Probably something like 50 is the lowest you can go without dropping a hundred or more elo, while 80 seems to produce more realistic evaluations in general than 100, though losing a couple dozen elo or so. On the upside, you might try to see how high you have to go to get a near-zero eval of the (Spanish) Marshall Gambit.

Re: To Larry Kaufman

Posted: Wed Mar 16, 2016 6:49 pm
by TShackel
lkaufman wrote:Well, the default is just 100% of what we believe makes for strongest play. The max is 200% (i.e. double), the min is zero, but if you set it to zero it will play silly, and even a setting like 1 or 2 will give ridiculous play. Maybe something like 10 is about the lowest value where it will play somewhat plausibly. Probably something like 50 is the lowest you can go without dropping a hundred or more elo, while 80 seems to produce more realistic evaluations in general than 100, though losing a couple dozen elo or so. On the upside, you might try to see how high you have to go to get a near-zero eval of the (Spanish) Marshall Gambit.
Thanks. I find this entire topic very interesting. Maybe I should be a programmer. As of right now I'm just a tester.. :)

This is my last question since I don't want to bug you too much. Were there any updates or re-works of contempt? And just to double check, the new contempt tries to avoid draws by repetition, avoid exchanges to some degree, and tries to avoid closed positions now since 9.3? Were there any more changes to Komodo 9.4's contempt.

I suppose you're going to answer all this when you release it though. So if you just want to wait until then that's fine.

Thanks again.

Sincerely,

Tim.

Re: To Larry Kaufman

Posted: Wed Mar 16, 2016 7:33 pm
by lkaufman
TShackel wrote:
lkaufman wrote:Well, the default is just 100% of what we believe makes for strongest play. The max is 200% (i.e. double), the min is zero, but if you set it to zero it will play silly, and even a setting like 1 or 2 will give ridiculous play. Maybe something like 10 is about the lowest value where it will play somewhat plausibly. Probably something like 50 is the lowest you can go without dropping a hundred or more elo, while 80 seems to produce more realistic evaluations in general than 100, though losing a couple dozen elo or so. On the upside, you might try to see how high you have to go to get a near-zero eval of the (Spanish) Marshall Gambit.
Thanks. I find this entire topic very interesting. Maybe I should be a programmer. As of right now I'm just a tester.. :)

This is my last question since I don't want to bug you too much. Were there any updates or re-works of contempt? And just to double check, the new contempt tries to avoid draws by repetition, avoid exchanges to some degree, and tries to avoid closed positions now since 9.3? Were there any more changes to Komodo 9.4's contempt.

I suppose you're going to answer all this when you release it though. So if you just want to wait until then that's fine.

Thanks again.

Sincerely,

Tim.
Comtempt is the same, as you describe above, but now it can be used in analysis mode with a new UCI option.

Re: To Larry Kaufman

Posted: Thu Mar 17, 2016 8:03 am
by shrapnel
Bump.
So when is the big day Komodo 9.4 will be released ?

Re: To Larry Kaufman

Posted: Thu Mar 17, 2016 8:35 am
by Nordlandia
Here is another request! Please make it so Komodo reset hash in percentage during each turn.

In Fritz and Arena Komodo stand at 99% from the early openings, i don't know why.