Is Komodo trying to become like Stockfish? To Larry kaufman

[TShackel]

Hi Larry,

I was just wondering if your current development of Komodo is based on trying to become like stockfish? I noticed with Komodo 9.0 there was a faster time to depth to some extent, and then with 9.1 there is an even bigger faster time to depth. So now the depths of Komodo are very comparable to stockfish. I'm wondering if this is a good idea.

I don't see any reason to try to get deeper and deeper narrower search when you were already a top program while having a wider search. I liked the fact that komodo is different from stockfish and can see more and perhaps miss fewer moves while stockfish may miss a few due to searching so deep and narrowly. I think rather than trying to make komodo the best komodo it can be, it seems you're trying to make it as much like stockfish as you can.

I just noticed that Komodo 9.1 was playing various positions worse than Komodo 9.0. I realize you play thousands of games to test changes, but it seems what I liked about Komodo 9.0 is kind of leaving in komodo 9.1 simply due to forcing komodo to search more deeply like stockfish. I hope you can understand the benefits of having a little bit wider ssearch in that you miss fewer moves. And you are proof that you can have a wider search than stockfish and still be a top program.

Anything you could do to build on the Komodo 9.0 release while retaining moderate (not excessively deep) search depths would be appreciated. i don't see a need to make Komodo search as deep as stockfish, I think there are other ways to improve Komodo 9.0's strength. Let me repeat, that I like what you did with Komodo 9.0 but I think Komodo 9.1 search depths are a little excessive, and was hoping you could think of other ways to improve the komodo 9.0 release.

Sincerely,

Tim.

P.S.- This is just constructive criticism. In now way am I forgetting to thank you and Mark for all the work you do on Komodo. I really liked Komodo 9.0 release but hope you can keep the development in the same spirit as Komdoo 9.0 with a little broader search and moderate search depths, instead of searching too deep just for its own sake.

[shrapnel]

What are you complaining about ?
Just change the Settings if you don't want Komodo 9.1 to reach great Depths too fast. Simple.
Komodo authors have given us full freedom to experiment with Settings, unlike rigid Stockfish authors.
Use the settings that suit you best.
I too am certainly NOT using Default settings !

[TShackel]

I'd rather get mark and larry to agree that the defaults should have moderate depths (rather than excessively deep) and wider search, rather than have to force myself to play around with settings which will likely just weaken their defaults. I'm a tester that usually uses the defaults of whatever the authors use for their latest version. I think it's unlikely the experts of komodo will be proven wrong by a casual user playing around with settings.

I'm not complaining nor am i unappreciative for their work. Komodo 9.0 was my favorite release of all the komodo's. I jjust think 9.1 took the depths to such an extreme that its compromising 9.0's good play. I was hoping they could find other ways to iimprove the elo and positional and attacking play of komodo, rather than try to mimick stockfishy depths and perhaps get rid of what made komodo 9.0 great.

Sincerley,

Tim.

[Laskos]

Hi Larry,

I was just wondering if your current development of Komodo is based on trying to become like stockfish? I noticed with Komodo 9.0 there was a faster time to depth to some extent, and then with 9.1 there is an even bigger faster time to depth. So now the depths of Komodo are very comparable to stockfish. I'm wondering if this is a good idea.

I don't see any reason to try to get deeper and deeper narrower search when you were already a top program while having a wider search. I liked the fact that komodo is different from stockfish and can see more and perhaps miss fewer moves while stockfish may miss a few due to searching so deep and narrowly. I think rather than trying to make komodo the best komodo it can be, it seems you're trying to make it as much like stockfish as you can.

I just noticed that Komodo 9.1 was playing various positions worse than Komodo 9.0. I realize you play thousands of games to test changes, but it seems what I liked about Komodo 9.0 is kind of leaving in komodo 9.1 simply due to forcing komodo to search more deeply like stockfish. I hope you can understand the benefits of having a little bit wider ssearch in that you miss fewer moves. And you are proof that you can have a wider search than stockfish and still be a top program.

Anything you could do to build on the Komodo 9.0 release while retaining moderate (not excessively deep) search depths would be appreciated. i don't see a need to make Komodo search as deep as stockfish, I think there are other ways to improve Komodo 9.0's strength. Let me repeat, that I like what you did with Komodo 9.0 but I think Komodo 9.1 search depths are a little excessive, and was hoping you could think of other ways to improve the komodo 9.0 release.

Sincerely,

Tim.

P.S.- This is just constructive criticism. In now way am I forgetting to thank you and Mark for all the work you do on Komodo. I really liked Komodo 9.0 release but hope you can keep the development in the same spirit as Komdoo 9.0 with a little broader search and moderate search depths, instead of searching too deep just for its own sake.

They are using the settings and time to depth to have better ELO, just that. In fact they achieved a lot of things: better single core scaling with increasing time control, better multi-core scaling. I guess the improvement from 9.0 to 9.1 in ELO could be 30 or more points on 16 cores at 40/40', and that's a lot considering increased draw rate. Stockfish would need to gain some 60 ELO points at the TC of testing framework in self-games because of magnification effect. Komodo now became a clear favorite to be TCEC champion once again.

[TShackel]

They are using the settings and time to depth to have better ELO, just that. In fact they achieved a lot of things: better single core scaling with increasing time control, better multi-core scaling. I guess the improvement from 9.0 to 9.1 in ELO could be 30 or more points on 16 cores at 40/40', and that's a lot considering increased draw rate. Stockfish would need to gain some 60 ELO points at the TC of testing framework in self-games because of magnification effect. Komodo now became a clear favorite to be TCEC champion once again.

What I'm saying is simple. Pick a different way to gain elo than what they did with 9.1....trying to become as deep as stockfish. Keep what made Komodo great seeing more in a wider search while still being the strongest program on the planet.

When it comes to "mindless" elo increases, I should remind you that Komodo according to the chessbase website strives to bring up its weak points, instead of increasing its strong point, ie tactics and open games. So I'd recommend Komodo tries to gain elo not by searching as deep as stockfish which completely changes the character of the komodo engine, but rather by closed position and attacking improvements (its weak points). Bring up the weak points and gain elo that way while keeping the nature of the komodo engine (searching a wider tree), instead of make komodo an even better open game tactical monster that can search 5 plies deeper while destroyiing what made komodo great.

Sincerely,

Tim.

[TShackel]

What are you complaining about ?
Just change the Settings if you don't want Komodo 9.1 to reach great Depths too fast. Simple.
Komodo authors have given us full freedom to experiment with Settings, unlike rigid Stockfish authors.
Use the settings that suit you best.
I too am certainly NOT using Default settings !

By the way, I don't like being accused of complaining (maybe i have a legitimate concern that i'm concerned about), and I also don't like you acting like you have to defend mark and larry on behalf of the attacker (ie me). I'm their biggest ally. But I don't like the direction they're heading from 9.0 (my favorite komodo release) to 9.1.

[mjlef]

Hi Larry,

I was just wondering if your current development of Komodo is based on trying to become like stockfish? I noticed with Komodo 9.0 there was a faster time to depth to some extent, and then with 9.1 there is an even bigger faster time to depth. So now the depths of Komodo are very comparable to stockfish. I'm wondering if this is a good idea.

I don't see any reason to try to get deeper and deeper narrower search when you were already a top program while having a wider search. I liked the fact that komodo is different from stockfish and can see more and perhaps miss fewer moves while stockfish may miss a few due to searching so deep and narrowly. I think rather than trying to make komodo the best komodo it can be, it seems you're trying to make it as much like stockfish as you can.

I just noticed that Komodo 9.1 was playing various positions worse than Komodo 9.0. I realize you play thousands of games to test changes, but it seems what I liked about Komodo 9.0 is kind of leaving in komodo 9.1 simply due to forcing komodo to search more deeply like stockfish. I hope you can understand the benefits of having a little bit wider ssearch in that you miss fewer moves. And you are proof that you can have a wider search than stockfish and still be a top program.

Anything you could do to build on the Komodo 9.0 release while retaining moderate (not excessively deep) search depths would be appreciated. i don't see a need to make Komodo search as deep as stockfish, I think there are other ways to improve Komodo 9.0's strength. Let me repeat, that I like what you did with Komodo 9.0 but I think Komodo 9.1 search depths are a little excessive, and was hoping you could think of other ways to improve the komodo 9.0 release.

Sincerely,

Tim.

P.S.- This is just constructive criticism. In now way am I forgetting to thank you and Mark for all the work you do on Komodo. I really liked Komodo 9.0 release but hope you can keep the development in the same spirit as Komdoo 9.0 with a little broader search and moderate search depths, instead of searching too deep just for its own sake.

Larry can give more, but my answer is that we are not trying to become more like any other program. The fact that Komodo is different is what makes it interesting. We are trying to make Komodo stronger (elo gain) by modifying its search and adding more knowledge to the evaluation. We have found some ways recently to increase the "search depth", and in the process make the program stronger. But Komodo still is not as aggressive as Stockfish in how it reduces parts of the search tree. Our testing has shown an elo gain from these changes so we keep them. Please note that all strong programs I know of reduce and prune a lot of the search tree. The root "depth" is really more of a search increment indicator. They are no longer full width programs. They can miss things. The secret is not to miss many important things. Compare the PV moves displayed by Stockfish and Komodo. You will generally see similar PV length, although Komodo's "depth" is lower. Komodo extends more moves and shape's its move tree differently. But either program can miss a lot. It is the elo gain that is important.

As for widening, that too is still very different. Some recent changes make Komodo a little less wide, and a little more deeper. Without going into what the changes were, rest assured they tested well and we think they make Komodo a stronger player and better chess analyst.

If you want less depth and more width, try adjusting Komodo's "Selectivity" and "Reduction" parameters. They help shape the search tree. Maybe you can find better settings than us and let us know what they are!

Mark

[Henk]

Yes you can play thousand games to test changes. But these games only have short time control otherwise it would take months to test one change.

So it still might be that searcher deeper is not a good strategy for games with long time control for they are pruning/reducing too much.

[TShackel]

Larry can give more, but my answer is that we are not trying to become more like any other program. The fact that Komodo is different is what makes it interesting. We are trying to make Komodo stronger (elo gain) by modifying its search and adding more knowledge to the evaluation. We have found some ways recently to increase the "search depth", and in the process make the program stronger. But Komodo still is not as aggressive as Stockfish in how it reduces parts of the search tree. Our testing has shown an elo gain from these changes so we keep them. Please note that all strong programs I know of reduce and prune a lot of the search tree. The root "depth" is really more of a search increment indicator. They are no longer full width programs. They can miss things. The secret is not to miss many important things. Compare the PV moves displayed by Stockfish and Komodo. You will generally see similar PV length, although Komodo's "depth" is lower. Komodo extends more moves and shape's its move tree differently. But either program can miss a lot. It is the elo gain that is important.

As for widening, that too is still very different. Some recent changes make Komodo a little less wide, and a little more deeper. Without going into what the changes were, rest assured they tested well and we think they make Komodo a stronger player and better chess analyst.

If you want less depth and more width, try adjusting Komodo's "Selectivity" and "Reduction" parameters. They help shape the search tree. Maybe you can find better settings than us and let us know what they are!

Mark

Hi Mark,

Thanks very much for your answer. I am relieved by your desire to keep the uniqueness of komodo present as you decide what changes to make. I want that very much first and foremost, but to also bring up komodo's weak points like closed positions and attack/pawn storming situations (which recent stockfish development versions still outplay komodo in).

I realize I'm not there with you when you guys decide what changes to make, but it just appeared to me from your previous comments on your site and on the forum that you were really fascinated with trying to get komodo to search with stockfish-like depths. I just thought that was a wrong motivation (to be like another program rather than improving what makes komodo great), when there are other good elo-gaining ways to improve komodo and keep it true to komodo's style.

Thanks for your comments. I asked for larry's opinion because I thoughth you were still traveling! But thanks for your reply as well.

Sincerley,

Tim.

[shrapnel]

If you want less depth and more width, try adjusting Komodo's "Selectivity" and "Reduction" parameters. They help shape the search tree. Maybe you can find better settings than us and let us know what they are!Mark

That's what I was trying to tell him, but he seems to have misunderstood.

I think it's unlikely the experts of komodo will be proven wrong by a casual user playing around with settings.

That's where you are wrong, Tim.
Even the Komodo Authors readily admit that its simply not possible for them to ascertain the best possible Settings for the various Parameters for all possible game and hardware conditions.
This is where 'casual' users like me sometimes prove useful to them.
I have found the Komodo authors very receptive and encouraging whenever I have communicated with them.