To Larry Kaufman

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
TShackel
Posts: 313
Joined: Fri Apr 04, 2014 10:09 pm
Location: Neenah, WI, United States

Re: To Larry Kaufman

Post by TShackel » Tue Mar 08, 2016 9:59 pm

Keep up the good work Larry! Looking forward to K 9.4 and the new parameter you think is interesting.

Sincerely,

Tim.

TShackel
Posts: 313
Joined: Fri Apr 04, 2014 10:09 pm
Location: Neenah, WI, United States

Re: To Larry Kaufman

Post by TShackel » Fri Mar 11, 2016 1:00 am

Hi Larry,

I just wanted to say I really liked Komodo 9.0 when it first came out because it seemed to sacrifice and understand compensation much more than all of its komodo predecessors. It sacrificed very frequently and won in amazing style. I noticed that 9.1, 9.2, and 9.3 have sacrificed far less than 9.0, although they still have occasionally here and there but with no surprise to the opponent.

My question is, what are some things you could do for a future version that might increase the dynamic, sacrificial play of komodo while at the same time remaining positionally sound that we're used to with komodo? To give examples of Komodo 9.0's sacrifices, it would sacrifice a pawn for a lead in development frequently, it would sacrifice the exchange for purely positional compensation, and it would win against tough opposition with that style. I have not seen many sacrifices since 9.0, only occasionally with a move that both engines typically see.

Any ideas why 9.0 might have had this ability and 9.1, 9.2, and 9.3 may have lost some of the speculative play?

Thanks for all your work. While I may wonder where the sacrifices are, there's no doubt in my mind you made komodo an overall stronger player with each release.

Sincerely,

Tim.

lkaufman
Posts: 3724
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: To Larry Kaufman

Post by lkaufman » Sat Mar 12, 2016 12:38 am

TShackel wrote:Hi Larry,

I just wanted to say I really liked Komodo 9.0 when it first came out because it seemed to sacrifice and understand compensation much more than all of its komodo predecessors. It sacrificed very frequently and won in amazing style. I noticed that 9.1, 9.2, and 9.3 have sacrificed far less than 9.0, although they still have occasionally here and there but with no surprise to the opponent.

My question is, what are some things you could do for a future version that might increase the dynamic, sacrificial play of komodo while at the same time remaining positionally sound that we're used to with komodo? To give examples of Komodo 9.0's sacrifices, it would sacrifice a pawn for a lead in development frequently, it would sacrifice the exchange for purely positional compensation, and it would win against tough opposition with that style. I have not seen many sacrifices since 9.0, only occasionally with a move that both engines typically see.

Any ideas why 9.0 might have had this ability and 9.1, 9.2, and 9.3 may have lost some of the speculative play?

Thanks for all your work. While I may wonder where the sacrifices are, there's no doubt in my mind you made komodo an overall stronger player with each release.

Sincerely,

Tim.
You'll be glad to hear that we are making this a UCI option in the next release. You can make Komodo play as wildly or as conservatively as you wish, with the default set for what we believe to give the strongest play. A low value means it will never sacrifice (without seeing the reward) while a high setting will make it gambit a pawn or the Exchange for just modest compensation.
Komodo rules!

TShackel
Posts: 313
Joined: Fri Apr 04, 2014 10:09 pm
Location: Neenah, WI, United States

Re: To Larry Kaufman

Post by TShackel » Sat Mar 12, 2016 1:01 am

lkaufman wrote:
TShackel wrote:Hi Larry,

I just wanted to say I really liked Komodo 9.0 when it first came out because it seemed to sacrifice and understand compensation much more than all of its komodo predecessors. It sacrificed very frequently and won in amazing style. I noticed that 9.1, 9.2, and 9.3 have sacrificed far less than 9.0, although they still have occasionally here and there but with no surprise to the opponent.

My question is, what are some things you could do for a future version that might increase the dynamic, sacrificial play of komodo while at the same time remaining positionally sound that we're used to with komodo? To give examples of Komodo 9.0's sacrifices, it would sacrifice a pawn for a lead in development frequently, it would sacrifice the exchange for purely positional compensation, and it would win against tough opposition with that style. I have not seen many sacrifices since 9.0, only occasionally with a move that both engines typically see.

Any ideas why 9.0 might have had this ability and 9.1, 9.2, and 9.3 may have lost some of the speculative play?

Thanks for all your work. While I may wonder where the sacrifices are, there's no doubt in my mind you made komodo an overall stronger player with each release.

Sincerely,

Tim.
You'll be glad to hear that we are making this a UCI option in the next release. You can make Komodo play as wildly or as conservatively as you wish, with the default set for what we believe to give the strongest play. A low value means it will never sacrifice (without seeing the reward) while a high setting will make it gambit a pawn or the Exchange for just modest compensation.
Wow Larry, that's the best news of the day! Great idea for a UCI option in that regard. It will be great for analysis too, because a higher value meaning more sacrifices will help komodo discover any hidden sacrifices hiding in a position for human GM's who want to analyze sharp positions.

That's great news again. Thanks much!

Sincerely,

Tim.

User avatar
Nordlandia
Posts: 2452
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: To Larry Kaufman

Post by Nordlandia » Sat Mar 12, 2016 3:00 pm

With that new parameter slightly adjusted I'm quite sure Komodo 9.4 will immediately detect 17 Rxb7!? from the famous Kasparov vs Shirov 1994 game.

[d]r2qk2r/1b3ppp/p2p1b2/2nNp3/1R2P3/2P5/1PN2PPP/3QKB1R w Kkq - 0 17

[d]r2qk2r/1n3ppp/p2p1b2/3Np3/1P2P3/2P5/2N2PPP/3QKB1R b Kkq - 0 18

Position is roughly in balance even full exchange down.

Maybe Larry know the necessary value in this case although he can't reveal it yet :wink:

User avatar
Leto
Posts: 2028
Joined: Thu May 04, 2006 1:40 am
Location: Dune

Re: To Larry Kaufman

Post by Leto » Sat Mar 12, 2016 3:28 pm

lkaufman wrote:
TShackel wrote:Hi Larry,

I just wanted to say I really liked Komodo 9.0 when it first came out because it seemed to sacrifice and understand compensation much more than all of its komodo predecessors. It sacrificed very frequently and won in amazing style. I noticed that 9.1, 9.2, and 9.3 have sacrificed far less than 9.0, although they still have occasionally here and there but with no surprise to the opponent.

My question is, what are some things you could do for a future version that might increase the dynamic, sacrificial play of komodo while at the same time remaining positionally sound that we're used to with komodo? To give examples of Komodo 9.0's sacrifices, it would sacrifice a pawn for a lead in development frequently, it would sacrifice the exchange for purely positional compensation, and it would win against tough opposition with that style. I have not seen many sacrifices since 9.0, only occasionally with a move that both engines typically see.

Any ideas why 9.0 might have had this ability and 9.1, 9.2, and 9.3 may have lost some of the speculative play?

Thanks for all your work. While I may wonder where the sacrifices are, there's no doubt in my mind you made komodo an overall stronger player with each release.

Sincerely,

Tim.
You'll be glad to hear that we are making this a UCI option in the next release. You can make Komodo play as wildly or as conservatively as you wish, with the default set for what we believe to give the strongest play. A low value means it will never sacrifice (without seeing the reward) while a high setting will make it gambit a pawn or the Exchange for just modest compensation.
That sort of sounds like Stockfish's contempt setting although the Komodo version sounds like a more extreme version which makes me very excited. At positive contempt Stockfish plays riskier lines, at negative contempt it is happy to draw. Stockfish's max setting of positive 100 still plays very strong and is perhaps less than 100 elo weaker than default. I've been wishing for a while for the Stockfish team to increase the max setting so that it could play even riskier chess, I mean why not have the max be 500 so that at positive 500 it'd play 300 or so elo weaker than default?

Larry/Mark can you make it so that once you set it in the parameters to make it play wildly it will do this in analysis mode as well rather than only doing so in play mode? That's one thing I dislike about Komodo 9.3's contempt setting, it only works in play mode, which makes it a hassle.

lkaufman
Posts: 3724
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: To Larry Kaufman

Post by lkaufman » Sat Mar 12, 2016 5:26 pm

Leto wrote:
lkaufman wrote:
TShackel wrote:Hi Larry,

I just wanted to say I really liked Komodo 9.0 when it first came out because it seemed to sacrifice and understand compensation much more than all of its komodo predecessors. It sacrificed very frequently and won in amazing style. I noticed that 9.1, 9.2, and 9.3 have sacrificed far less than 9.0, although they still have occasionally here and there but with no surprise to the opponent.

My question is, what are some things you could do for a future version that might increase the dynamic, sacrificial play of komodo while at the same time remaining positionally sound that we're used to with komodo? To give examples of Komodo 9.0's sacrifices, it would sacrifice a pawn for a lead in development frequently, it would sacrifice the exchange for purely positional compensation, and it would win against tough opposition with that style. I have not seen many sacrifices since 9.0, only occasionally with a move that both engines typically see.

Any ideas why 9.0 might have had this ability and 9.1, 9.2, and 9.3 may have lost some of the speculative play?

Thanks for all your work. While I may wonder where the sacrifices are, there's no doubt in my mind you made komodo an overall stronger player with each release.

Sincerely,

Tim.
You'll be glad to hear that we are making this a UCI option in the next release. You can make Komodo play as wildly or as conservatively as you wish, with the default set for what we believe to give the strongest play. A low value means it will never sacrifice (without seeing the reward) while a high setting will make it gambit a pawn or the Exchange for just modest compensation.
That sort of sounds like Stockfish's contempt setting although the Komodo version sounds like a more extreme version which makes me very excited. At positive contempt Stockfish plays riskier lines, at negative contempt it is happy to draw. Stockfish's max setting of positive 100 still plays very strong and is perhaps less than 100 elo weaker than default. I've been wishing for a while for the Stockfish team to increase the max setting so that it could play even riskier chess, I mean why not have the max be 500 so that at positive 500 it'd play 300 or so elo weaker than default?

Larry/Mark can you make it so that once you set it in the parameters to make it play wildly it will do this in analysis mode as well rather than only doing so in play mode? That's one thing I dislike about Komodo 9.3's contempt setting, it only works in play mode, which makes it a hassle.
The next release of Komodo will have an option for contempt in analysis mode. But contempt has nothing to do with the new term. It controls the relative weight of dynamic vs. static features, nothing to do with playing for a draw or a win. A high setting emphasizes things like mobility, king safety, etc., while a low setting emphasizes material and pawn structure.
Komodo rules!

TShackel
Posts: 313
Joined: Fri Apr 04, 2014 10:09 pm
Location: Neenah, WI, United States

Re: To Larry Kaufman

Post by TShackel » Sat Mar 12, 2016 5:59 pm

lkaufman wrote:The next release of Komodo will have an option for contempt in analysis mode. But contempt has nothing to do with the new term. It controls the relative weight of dynamic vs. static features, nothing to do with playing for a draw or a win. A high setting emphasizes things like mobility, king safety, etc., while a low setting emphasizes material and pawn structure.
The new term sounds great Larry. I like the option of a more dynamic komodo. Does your default setting that you find to be strongest improve the dynamic abililty of komodo to some degree too? Or will only a high value improve sacrifices or dynamics which will not be in the default? Because it would be nice to find a weight to make komodo a bit more dynamic with some kind of a high value but obviously paying the most attention to elo gain.

To take this a step further, I wonder if it's possible to set up a higher positive value for openings or middlegames, and a lower value for endgames since static factors would be more important in endgame. I hope a dynamic komdoo will not burn its own britches by doubling pawns around kingside which is clearly a move that a GM would never make in order to make itself more dynamic at all costs. So I suppose its important to weight static factors in opening and middlegame to some extent too, as that will affect the endgame if it makes a big structural mistake early in the game. So I suppose this term needs a delicate balance in all three phases of a chess game.

My goal obviously is to have Komodo be the strongest player possible and I wouldn't like to see unsound sacrifices that make komodo lose. However, Komodo 9.0 was great because it would play these sacrifices and win against strong opposition! So it sacrificed and was strongest! If that can happen it would be amazing.

Keep up the good innovative work.

Sincerely,

Tim.

lkaufman
Posts: 3724
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: To Larry Kaufman

Post by lkaufman » Sat Mar 12, 2016 6:47 pm

TShackel wrote:
lkaufman wrote:The next release of Komodo will have an option for contempt in analysis mode. But contempt has nothing to do with the new term. It controls the relative weight of dynamic vs. static features, nothing to do with playing for a draw or a win. A high setting emphasizes things like mobility, king safety, etc., while a low setting emphasizes material and pawn structure.
The new term sounds great Larry. I like the option of a more dynamic komodo. Does your default setting that you find to be strongest improve the dynamic abililty of komodo to some degree too? Or will only a high value improve sacrifices or dynamics which will not be in the default? Because it would be nice to find a weight to make komodo a bit more dynamic with some kind of a high value but obviously paying the most attention to elo gain.

To take this a step further, I wonder if it's possible to set up a higher positive value for openings or middlegames, and a lower value for endgames since static factors would be more important in endgame. I hope a dynamic komdoo will not burn its own britches by doubling pawns around kingside which is clearly a move that a GM would never make in order to make itself more dynamic at all costs. So I suppose its important to weight static factors in opening and middlegame to some extent too, as that will affect the endgame if it makes a big structural mistake early in the game. So I suppose this term needs a delicate balance in all three phases of a chess game.

My goal obviously is to have Komodo be the strongest player possible and I wouldn't like to see unsound sacrifices that make komodo lose. However, Komodo 9.0 was great because it would play these sacrifices and win against strong opposition! So it sacrificed and was strongest! If that can happen it would be amazing.

Keep up the good innovative work.

Sincerely,

Tim.
The new term cannot be set differently for middlegame or endgame, but I think there is no need. If a move allows the opponent more checks or attacks on our king, a higher setting will avoid the move. In other words, it's symmetric; it treats both sides the same.
We could be off a few percent in the optimum setting, and it may also be that the optimum setting is quite different for long time control games than for the levels at which we can test. I suppose if you set it for something like 110% of our recommendation the elo loss will be pretty small, probably single digit, so if that makes Komodo aggressive enough for you you can use that all the time (except for official tests of course). But a really high setting like 150% would cost a lot of elo. Pawns do have value!
Komodo rules!

User avatar
Nordlandia
Posts: 2452
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: To Larry Kaufman

Post by Nordlandia » Sat Mar 12, 2016 9:35 pm

Someone suggested adjustable 50-move rule for certain positions that is usually a draw, e.g Behting Study.

For example 25-move rule may be adequate in that particular position.

Is this possible to add in Komodo parameters?

https://chessprogramming.wikispaces.com/Behting+Study

Post Reply