Komodo 12 and MCTS

[mjlef]

I am not sure what they mean by "same analysis technology" as..... If it is a neural network, then no, are are not doing that right now. If it means Monte Carlo Tree Search, then yes we are using that. I did not get a copy of this text until I saw it here. I did ask that some other copy which claims Komodo plays thousands of games in MCTS mode be corrected, since we (and Leela and Alpha Zero) are not doing that. They did fix the text for this second thing. We will look into this.

Thanks for bringing this to our attention. I want it to be as accurate as possible.

[Leo]

Otherwise they'll be selling an early prototype that will soon become outdated.

Pointless having MCTS option when its unable to use the power of the GPU or of Tensor Processing Units.
Sorry, but the simple truth is that this is just a Sales Gimmick by the Komodo team trying to take advantage of the current interest in MCTS and Neural Networks.

Its not a gimmick. They are using something new. It might revolutionize the way Komodo runs. You are upset and acting selfish because there is nothing in it for you.

[glennsamuel32]

I am not sure what they mean by "same analysis technology" as..... If it is a nueral network, then no, are are not doing that right now. If it means Monte Carlo Tree Search, then yes we are using that. I did not get a copy of this text until I saw it here. I did ask that some other copy which claims Komodo plays thousands of games in MCTS mode be corrected, since we (and Leela and Alpha Zero) are not doing that. They did fix the text for this second thing. We will look into this.

Thanks for bringing this to our attention. I want it to be as accurate as possible.

I have no doubt that this is commercial promotion from Chessbase for it's products -- just like Honda promoting it's VTECH TURBO as the very future of car engines !!

[matejst]

I am not a Komodo customer and, living where I live and earning as little as I earn, so I want to express my gratitude to Larry and Mark for offering the previous versions of their great engine, an engine that doesn't need expensive hardware to work, but plays some exquisite chess on my 10 years old laptop.

I wish also to add that I am sure that the paradigm with HI behind engine play has still a lot to give, and that relying on books and TBs was perhaps the wrong way to go, or the wrong big data to use. A lot of knowledge of different nature can be added to engines, which will show its value with time.

Once again, thanks Larry, thanks Mark!

[Milos]

Its not a gimmick. They are using something new. It might revolutionize the way Komodo runs. You are upset and acting selfish because there is nothing in it for you.

It is not gonna revolutionize anything. It is easy to plot scaling of Komodo MCTS vs regular Komodo and see that one would need at least 1000x more cores than what we have today (like 8000 instead of 8) for Komodo MCTS to surpass regular Komodo, which with current pace of technology scaling that is drastically reduced would happen at best in 50 years.

[Uri Blass]

I would love someone to try another experiment. Take the Leela neural network, and just write a classic search for it, using the Leela neural network in place of an eval, but include all the standard cutoffs. If easier, you could convert the win percentage back to a centipawn score, and try all the standard pruning tricks we do in programs like Stockfish. I would love to see how it performs. Neural networks can in many ways see future moves via piece cooperation, attack and defense terms. It might be a very powerful program.

Interesting idea. I thought about this before

Replacing the regular eval with the Leela NN probably won't work, the NN is ~1000 times slower and this will wreck your engine tactically. Maybe there is a way to make some kind of hybrid, using the NN closer to the root for pruning decisions and using the classic search for the remaining things.

I think the right comparison is against Leela that is already weak in tactics and the question is if normal search can help leela to play better.

I also think that speed is not everything and if you have a better evaluation then you may earn more from pruning.

[peritodeifinali]

[/quote]

will there be updates for this version?

[main line]

Very long infinite analysis won't even work with MCTS on right now, it will just stop analyzing when reaching memory limits set by the program (not by user). We don't have any basis for recommending a value other than the default for longer time contnrols or analysis; the MCTS mode is just too new. Don't bother setting Skill for anything other than default 20 with MCTS on right now, it will only have small and somewhat unpredictable effects. Most of the UCI options just don't work yet with MCTS on, it's just too new. Syzygy also doesn't work yet with MCTS on. Hash size, table memory, Ponder, and Threads (up to 3) should work ok with MCTS. We'll gradually make more UCI options work properly with MCTS over time. Threads beyond 3 is top priority.

Look at this


Monte Carlo can't find Qc4!! played by AlphaZero. And Monte Carlo analysis doesn't work

[AdminX]

It spots on my system.

[Ozymandias]

Can you try with MPV off?