Can somebody explain what makes Leela as strong as Komodo?

[grahamj]

Is it the Hardware speed or some breakthrough programming innovation?
So far what is the score between Komodo and Leela ?

My answer is a bit different than others. Leela (following AlphaZero) has found a way to use a powerful GPU for chess, which is itself composed of hundreds or even thousands of cores. So far the only method that has made use of the GPU for chess is Neural Network. If we (Komodo or any other non-NN engine) could find a way to make use of these thousands of cores for chess other than NN we would probably have clearly the top engine, but it's not easy because these GPU cores are not able to do all the things a CPU can do. If Leela (or AlphaZero) runs on a CPU, it's rather pathetic.

There’s a chess engine reported a few years ago that runs on gpu. Gpu has enough instructions to also operate as a “computer” and, as you presumably know, one can use logic operations as a alternative to branching (branching instructions exist I think, but are not so easy as with cpu.)

See

http://chessgpgpu.blogspot.com/

and

http://nikolachess.com/

It is a pity that the links of the videos don't work, only the donation button works!

I think those were bogus, but Zeta OpenCL was the real deal. Not a powerhouse, because it was not a NN design, but it played decent chess.
I think GPUs will be a natural for mate proof search because you only need win/loss/draw information and the fastest move generator in the world runs on GPUs.

http://chessgpgpu.blogspot.com/ is not bogus. I learned quite a bit from reading that blog. Perhaps mostly what not to do, but the author is serious.

http://nikolachess.com/ on the other hand looks utterly absurd. 'The goal is 16000 ELO and 32 men-tablebase completed one day. ... Donate here.'

[Leo]

The Nikola site has been online for years. They tried for a $500,000 grant to rent a supercomputer. As far as I have heard, it literally is impossible in 2019 to store all the moves of a 32 man TB.

[Chessqueen]

Is it the Hardware speed or some breakthrough programming innovation?
So far what is the score between Komodo and Leela ?

I'll take a quick stab and also try to keep it simple to understand the gist without getting mired down in all the nitty gritty details.

You might have heard phrase Neural Network or "NN". The theory of NN has been around for years - check this summer thesis submitted by Justin Boyan in 1992:

Modular Neural Networks for Learning Context􏰆 Dependent Game Strategies
http://citeseerx.ist.psu.edu/viewdoc/do ... 1&type=pdf

NN been round for over 30 years - but it's only been recently ( last ten years) where the hardware ( and the software written to take advantage of the hardware) has taken this to a much higher level than playing a game. Neural Network and its related cousin Data Analytics are huge today at major companies thought out the world - not for playing games - but for making real time business decisions. A perfect example is that my former company developed a model to predict which of their customers ( over a hundred million) will mostly likely default and use that as to how much credit will be extended and the type of remedies they will use when they do default to maximize their recovery (they called it Predictive Analytics) - and it's all automated . Google, Amazon , Facebook etc are all into this big time and they probably know you better than your friends know you.

How does Leela work - it plays scores of millions of games with itself and analyze the moves , the position, and the result of those games to come up with a predictive analytic outcome for each position. It takes the the position that has the highest chance of winning the game based on its history and analysis of its passed games and plays that move. In fact, Leela has an output option to display White or Black's "chances of winning percentage" rather than the customary centipawn evaluation. This is a gross simplification of the process - to gain a better and more compete understanding - read through this link below which has links to 20 or so other websites with more information:

https://deepmind.com/blog/alphazero-she ... gi-and-go/

I read the Modular Neural Network and how is suppose to work, and my understanding is that Komodo or SF are still stronger based on comparable hardware. But I do NOT know if they made a version of Leela or Alpha Zero that can work on regular PC purchase off the shelf? And if NOT Komodo and SF 10 are still stronger for us the regular users

[Nay Lin Tun]

@chessqueen, It is already known that Leela on high end gamer pc beat SF dev in any fast time control <30 min, rapid/blitz/bullet time controls , 2080 Leela vs 4 cores SF. (GPU favours in gamer's PC )

You can setup Leela in your PC anytime. Leela is available to all consumers for free. (To play decent standard of Leela , your minimum requirement is $250 ,1060 GTX)

And those two programs will be in superfinal though.

[Dann Corbit]

The Nikola site has been online for years. They tried for a $500,000 grant to rent a supercomputer. As far as I have heard, it literally is impossible in 2019 to store all the moves of a 32 man TB.

According to one estimate: https://tromp.github.io/chess/chess.html

There are about 45193640626062205213735739171550309047984050718 chess positions, so I guess that storing them will be kind of expensive.

I think he's going to need a bigger grant.

Generally speaking, those estimates do not take promotions, half move clock or repetition into account.
Half move clock and repetition can be independently calculated, so I can see where that part makes sense. But promotions are pretty important.

But will:
KQQQQQQQQQRRBBNNvk be a big file or a little one?
Either way, I think k is in trouble.

[reflectionofpower]

As the street speaketh,"It is what it is."

[jp]

Like lkaufman, I think it's mainly a successful exploitation of hardware.

I look at the AlphaZero paper in two quite different ways. As a scientific contribution, the important thing is the generality of the algorithm. The importance ... is to demonstrate the method can work when little human knowledge is available. Many important real-world inference problems are poorly understood by humans.

On the other hand, from the very narrow perspective of chess engines, I do not think AlphaZero represents the future.

The generality of the algorithm is greatly overestimated. It would be general if no work at all was required to represent the game of chess & make it all work. Obviously that's not true.

It does not have much connection to "important real-world inference problems" either.

[Chessqueen]

Do you mean Leela, in general, or Leela today vs. 6 months ago?

This is the most recent match that I found:

[dkappe]

If you want some smaller leela nets that run better on PC, you can check out some of the 1CPU tests running under CCRL 40/40 conditions. http://talkchess.com/forum3/viewtopic.php?f=6&t=69633

[Uri Blass]

Leela, lcz wins because it has no AB search.

It uses a different search and a different evaluation. That makes it difficult for AB programs to find out what it is planning.

It's like chess in a parallel universe.
It's very difficult to find out for the AB animal.

You can call it paradigm change

But the true next step in computer chess development is the moment when the machine is planning something.
So far the machines find a bunch of good moves by chance. The AB programs do not really play chess. They solve test suites.
The next step is that the engine defines a plan and tries to make this plan work. By using moves.

Not having AB search is not the reason that lcz wins.
Give lcz the same cpu as AB engines and lcz is going to lose.

lcz wins because it use new hardware that AB engines are not designed to use.
it does not mean that it is impossible to make effective use of the new hardware by AB engines.

Maybe we are going to see better AB engines that use the GPU in the future.