Komodo 12.1

[mjlef]

My understanding is that komodo MCT is a derivative of classic Komodo.

I do not understand how people can think different when probably more than 90% of the chess code is the same.
Note that I have no problem if tournaments allow derivatives to compete.

I completely agree. Over time we might find we need to change the eval and search a lot and they might begin to differ more over time. But they certainly will remain derivatives without some massive total program rewrite. In the Leela case, I believe they use Stockfish code for board representation, and move generation, but the "evaluation: is from the neural network. More different than Komodo to Komodo MCTS, but much of the code is the same so it is also a derivative. It would be nice to establish rules to determine when two programs are too similar to be in the same tournament. Is a vastly different evaluation enough?

You are saying that sharing the move generator and board representation makes it a clone? Even if the search and eval are all so different as to be incomparable?

I did not say "clone" which is a vaguely defined term. I said "derivative". Strength is gained in a chess program of it has an efficient board representation and fast move generator. And saves a programmer a lot of time from writing it from scratch. Since MCTS in takes most of the time, I would be fine to have it in the same tournament as Stockfish. But what is permitted would depend on the rules of the specific tournament. I just think it would be good to get consensus on where the dividing line should be.

[Ron Langeveld]

Oh. I thought one of the great hopes of this approach was its potential on massive hardware. If it doesn’t scale better than alpha beta >12 cores...the old way will always win as 12 cores will be increasingly common over the next few years.

Just a few posts back Larry did write "I expect we'll find a way to use more than 12 threads before too long."

I suggest you don't develop an accidental blindspot to free info by developers (in general) that tend to be reluctant with info on future versions. You may have misread it, but from where I am sitting I applaud the Komodo team for listening to customers and providing a significant enhancement within a couple of weeks. If you are one of the few Komodo customers with a core count that far exceeds the 12 core boundary for now than posts like this won't get any pity from me

Ron

Missing the point entirely, nothing to do with my hardware but the future of the project and TCEC

You jump to conclusions like 'the old way will always win' without bothering to read the available post from Larry and then think that I am missing something by mentioning 'the future of the project and TCEC' for the very first time in this thread

Such absence of any consistency in argumentation and logic is symptomatic for one thing only: trolling.

Welcome to my banlist!

Ron

[Albert Silver]

I completely agree. Over time we might find we need to change the eval and search a lot and they might begin to differ more over time. But they certainly will remain derivatives without some massive total program rewrite. In the Leela case, I believe they use Stockfish code for board representation, and move generation, but the "evaluation: is from the neural network. More different than Komodo to Komodo MCTS, but much of the code is the same so it is also a derivative. It would be nice to establish rules to determine when two programs are too similar to be in the same tournament. Is a vastly different evaluation enough?

You are saying that sharing the move generator and board representation makes it a clone? Even if the search and eval are all so different as to be incomparable?

I did not say "clone" which is a vaguely defined term. I said "derivative". Strength is gained in a chess program of it has an efficient board representation and fast move generator. And saves a programmer a lot of time from writing it from scratch. Since MCTS in takes most of the time, I would be fine to have it in the same tournament as Stockfish. But what is permitted would depend on the rules of the specific tournament. I just think it would be good to get consensus on where the dividing line should be.

Actually, in LC0 the board representation was changed and is more efficient according to Alexander Lyaschuk, the developer. As to the move generator, I am sure you are not wrong, but suspect that with an engine that has trouble producing even 2000 nodes per second, the effect is going to be a lot less than with one that produces millions.

[mjlef]

You are saying that sharing the move generator and board representation makes it a clone? Even if the search and eval are all so different as to be incomparable?

I did not say "clone" which is a vaguely defined term. I said "derivative". Strength is gained in a chess program of it has an efficient board representation and fast move generator. And saves a programmer a lot of time from writing it from scratch. Since MCTS in takes most of the time, I would be fine to have it in the same tournament as Stockfish. But what is permitted would depend on the rules of the specific tournament. I just think it would be good to get consensus on where the dividing line should be.

Actually, in LC0 the board representation was changed and is more efficient according to Alexander Lyaschuk, the developer. As to the move generator, I am sure you are not wrong, but suspect that with an engine that has trouble producing even 2000 nodes per second, the effect is going to be a lot less than with one that produces millions.

]]

I stand corrected. I actually looked over the source for the move gen, position and board stuff and I could find no difference other than things removed that are no longer needed (like move ordering), but maybe I missed something. There is even a quiet checking move generator I do not see being used by Leela at all. If you know the differences, let me know. I am old and miss things. Anyway, it is a derivative, but one many tournaments would probably accept since the search and eval are massively different.

[Werewolf]

[
You jump to conclusions like 'the old way will always win'

No I didn't, but you cannot read (when did I ever say it was about "my hardware"), think or understand computer chess so there's no point talking to you

[Nordlandia]

Latest K vs H6 with correct contempt values, will be a neat match to watch.

[CMCanavessi]

Latest K vs H6 with correct contempt values, will be a neat match to watch.

Define "correct"

[shrapnel]

Komodo 12.1 is planned for release this morning at komodochess.com. While the base program has only one minor bugfix from Komodo 12, perhaps worth two or three elo, the MCTS version has been massively revised and improved. Estimated Elo gain on one, two, or three threads about 100 points, and since the thread limit has been raised from three to twelve the gain grows even larger with each added thread. "MCTS Hash" has been added so users with lots of cores and memory can allow MCTS to run for hours instead of minutes per position before filling up the Hash. Testing groups are encouraged to test the MCTS version, especially on four (or even up to 12) threads, as we believe it will gain more than most normal engines from MP.

No matter how many improvements you plan for the MCTS version, NOTHING can change the FACT that Komodo MCTS, without NN, is still standing on one leg.
You can beautify that leg, put armor on it or make whatever Plans you want, but its still standing on one leg.
Komodo can NEVER hope to match AlphaZero so long as it is unable to use Neural Networks and thereby the Power of the GPU, never mind the fancy terms/advertisements.
By the way, I seem to remember that barely a year ago, I was laughed off the Forum when I suggested that Chess Engines could never really increase in strength unless they reduced their dependence on the CPU and learned to utilize the amazing Power of the GPUs.
People very condescendingly explained to me that such a thing was impossible and could never happen and I didn't know what I was talking about.
Well, NOW it appears that nearly every conventional chess engine is scrambling to try to catch up with AlphaZero, after the latter simply blew away Stockfish . What REALLY got the collective Goats of the old Sticks-in-the-Mud here was the CONTEMPTUOUS ease with which it did so !
Many of them are still in Denial mode !
Anyway, it looks like I had the last laugh, after all !

[Nordlandia]

Latest K vs H6 with correct contempt values, will be a neat match to watch.

Define "correct"

Contempt = zero for both parties.

[lkaufman]

Komodo 12.1 is planned for release this morning at komodochess.com. While the base program has only one minor bugfix from Komodo 12, perhaps worth two or three elo, the MCTS version has been massively revised and improved. Estimated Elo gain on one, two, or three threads about 100 points, and since the thread limit has been raised from three to twelve the gain grows even larger with each added thread. "MCTS Hash" has been added so users with lots of cores and memory can allow MCTS to run for hours instead of minutes per position before filling up the Hash. Testing groups are encouraged to test the MCTS version, especially on four (or even up to 12) threads, as we believe it will gain more than most normal engines from MP.

No matter how many improvements you plan for the MCTS version, NOTHING can change the FACT that Komodo MCTS, without NN, is still standing on one leg.
You can beautify that leg, put armor on it or make whatever Plans you want, but its still standing on one leg.
Komodo can NEVER hope to match AlphaZero so long as it is unable to use Neural Networks and thereby the Power of the GPU, never mind the fancy terms/advertisements.
By the way, I seem to remember that barely a year ago, I was laughed off the Forum when I suggested that Chess Engines could never really increase in strength unless they reduced their dependence on the CPU and learned to utilize the amazing Power of the GPUs.
People very condescendingly explained to me that such a thing was impossible and could never happen and I didn't know what I was talking about.
Well, NOW it appears that nearly every conventional chess engine is scrambling to try to catch up with AlphaZero, after the latter simply blew away Stockfish . What REALLY got the collective Goats of the old Sticks-in-the-Mud here was the CONTEMPTUOUS ease with which it did so !
Many of them are still in Denial mode !
Anyway, it looks like I had the last laugh, after all !

You may be right, but just saying so doesn't make it so. It is not at all clear that the AlphaZero success was due to the NN; it may have been due primarily to a good implementation of MCTS. If we can raise Komodo MCTS to be competitive with normal Komodo at something like 3 minutes per move without using GPU, that would indicate that MCTS was perhaps the key. But we aren't there yet. We may well try NN and GPUs, but personally I'm not convinced that it will help. There are many things we can do to improve MCTS besides using NN. We haven't even implemented any intelligent time management yet, for example.