The next revolution in computer chess?

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

Moderators: hgm, Rebel, chrisw

User avatar
Ovyron
Posts: 4557
Joined: Tue Jul 03, 2007 4:30 am

Re: The next revolution in computer chess?

Post by Ovyron »

chrisw wrote: Fri Jul 24, 2020 8:53 pm Would you like to buy this dotcom business I have to sell, it’s been going up and up and up and up and everyone can see where this is going (up).
Can it go down? Because businesses go down all the time. NNUE nets only get stronger with more training.

Businesses have nothing to do with chess engines, as you can see how Stockfish dev's elo continues to go up despite chess' exponential nature.
chrisw wrote: Fri Jul 24, 2020 8:53 pmUnless it’s an orangutan that is almost fully developed brain wise and won’t learn to read and write however hard you try.
Except NNUE already reads and writes better than any other chess entity in the world. It's already the best and it hasn't even been properly trained.

What you're missing is that EVERYONE ELSE IS ALREADY TRYING TO CATCH UP TO NNUE, it has won the best approach to chess, you got lost looking at a tree, need to zoom out a bit to see the whole forest.
chrisw
Posts: 4457
Joined: Tue Apr 03, 2012 4:28 pm
Location: Midi-Pyrénées
Full name: Christopher Whittington

Re: The next revolution in computer chess?

Post by chrisw »

Ovyron wrote: Fri Jul 24, 2020 9:08 pm
chrisw wrote: Fri Jul 24, 2020 8:53 pm Would you like to buy this dotcom business I have to sell, it’s been going up and up and up and up and everyone can see where this is going (up).
Can it go down? Because businesses go down all the time. NNUE nets only get stronger with more training.

Businesses have nothing to do with chess engines, as you can see how Stockfish dev's elo continues to go up despite chess' exponential nature.
chrisw wrote: Fri Jul 24, 2020 8:53 pmUnless it’s an orangutan that is almost fully developed brain wise and won’t learn to read and write however hard you try.
Except NNUE already reads and writes better than any other chess entity in the world. It's already the best and it hasn't even been properly trained.

What you're missing is that EVERYONE ELSE IS ALREADY TRYING TO CATCH UP TO NNUE, it has won the best approach to chess, you got lost looking at a tree, need to zoom out a bit to see the whole forest.
Yes, thanks for the upper case, but it doesn’t answer the point. Because something is on a rising trend does not mean that trend is going to linearly continue. Ironic everybody jumping up and down predicting a linear and endless improvement trend when the entity showing the trend absolutely demonstrates the power of non-linearity. All credit to the guy who came up with the smart tricks to render possible the time-shrinking of chess and other games neural nets. Beyond that there’s nothing new in maybe thirty years. It’s just a standard well known fully connected net which can output some approximation to a chess eval fast enough to compete.
Dann Corbit
Posts: 12662
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: The next revolution in computer chess?

Post by Dann Corbit »

smatovic wrote: Fri Jul 24, 2020 5:55 pm
corres wrote: Fri Jul 24, 2020 5:47 pm
Rebel wrote: Fri Jul 24, 2020 4:55 pm
corres wrote: Fri Jul 24, 2020 1:48 pm
Rebel wrote: Fri Jul 24, 2020 11:20 am ...
NNUE engines don't need expensive GPU cards.
This is the only one real benefit.
Sure, everybody is wrong and you are right.
The future will decide about who is wrong and who is right.
In every cases I keep my NVIDIA cards.
Hehe, dude, the future is now :)

http://ccrl.chessdom.com/ccrl/404/

Not sure if a such a thing happened ever before, #1 top entry on its first run @CCRL...

better invest in some AVX based CPU too ;-)

--
Srdja
Pohl saw the same:
https://www.sp-cc.de/
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Dann Corbit
Posts: 12662
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: The next revolution in computer chess?

Post by Dann Corbit »

chrisw wrote: Fri Jul 24, 2020 8:53 pm Unless it’s an orangutan that is almost fully developed brain wise and won’t learn to read and write however hard you try.
It does not refute your point (which was obviously metaphorical anyway) but:
https://en.wikipedia.org/wiki/Chantek

Not only that but:
https://www.bbcearth.com/blog/?article= ... %20belongs.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
chrisw
Posts: 4457
Joined: Tue Apr 03, 2012 4:28 pm
Location: Midi-Pyrénées
Full name: Christopher Whittington

Re: The next revolution in computer chess?

Post by chrisw »

Dann Corbit wrote: Fri Jul 24, 2020 9:46 pm
chrisw wrote: Fri Jul 24, 2020 8:53 pm Unless it’s an orangutan that is almost fully developed brain wise and won’t learn to read and write however hard you try.
It does not refute your point (which was obviously metaphorical anyway) but:
https://en.wikipedia.org/wiki/Chantek

Not only that but:
https://www.bbcearth.com/blog/?article= ... %20belongs.
Yes, the orangutan was the usual hook. Asymptote would never have reached the target audience.
User avatar
towforce
Posts: 11858
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK

Re: The next revolution in computer chess?

Post by towforce »

chrisw wrote: Fri Jul 24, 2020 7:52 pmYes, else ply 0 Leela would beat GMs at non-blitz and GMs/IMs can mostly express in language why so and so move and so on.

I would agree on this point: I want chess solved before I die. If nobody else will do it then I will.

The results are very impressive, but it comes at a price: the NN, and hence the amount of knowledge it can store, has been made smaller. From the POV of solving chess, this is a slight negative.

The main problem for getting in the non-linearities is already solved by tuning.

No. You can't solve chess by taking a numerical expression and solving its weightings. You need an expression that captures everything that needs to be captured, and THEN you need to tune it.

The difference between tuning a polynomial of known themes and a layered spreadsheet of one hots, is that human can understand the polynomial and devise more terms and combine terms, all on basis of human ingenuity...

Not going to happen: the chess polynomial will not be a common 2 variable polynomial (y = 4x^2 - 2x - 3).

Chess has:

* 64 squares
* 7 pieces (rook, knight, bishop (ws), bishop(bs), queen, king, pawn)
* 2 colours

So the base number of variables in the polynomial will be: 64*7*2 = 896

Each of the bishops can only go on half the squares, so reduce that by 4 * 32 = 128

There are 16 squares that the 16 pawns can't go on, so reduce the number again by 16 * 16 = 256

So the number of polynomial variables will be 896 - 256 - 128 = 512

That's too many dimensions for the human brain to work in, unless somebody can come up with a way of visualising a polynomial in that many dimensions.

Could I write a program to fit curves in 512 dimensions? HELL YES!!! :twisted:

Worse still, I can't think of anything else I'd rather be doing!

I'm not going into detail in this thread (this is NNUE's place), but I'll quickly say two things:

1. NN's don't require accurate arithmetic. This is why vector processing chips for NN training (like Google's TPU - link) tend to use half-precision arithmetic. In respect of this, to make my curve fitting easier, I'm planning to have the input data in ranges rather than exact points. This will make the curve fitting MASSIVELY easier, and keep the polynomial level as low as possible (high level polynomials are lumpy, low level polynomials are smooth. Smoother is better!)

2. In contrast to NN learning, this curve-fitting does have one cost: in the input data, you can never have the same same chess position twice with different "scores" (or incompatible score ranges), so the data will have to be checked for this before the curve fitting begins.

Thank you for the spreadsheet mention. I looked it up, and while it might be difficult to train an NN in a spreadsheet, a trained NN can certainly be written as a spreadsheet. There's an article here - I've already copied the sheet into my own documents!

, whereas the layered spreadsheet technician has absolutely no idea at all what is going on (well maybe some heat maps, but those are in effect no more than pretty pictures). Spreadsheet technician is at the mercy of multi-dimensional connections that make no readable sense.
The simple reveals itself after the complex has been exhausted.
matejst
Posts: 366
Joined: Mon May 14, 2007 8:20 pm
Full name: Boban Stanojević

Re: The next revolution in computer chess?

Post by matejst »

Uly,

sorry, I guess I misunderstood you -- the written word is treacherous.
Albert Silver
Posts: 3026
Joined: Wed Mar 08, 2006 9:57 pm
Location: Rio de Janeiro, Brazil

Re: The next revolution in computer chess?

Post by Albert Silver »

towforce wrote: Fri Jul 24, 2020 11:37 pm
chrisw wrote: Fri Jul 24, 2020 7:52 pmYes, else ply 0 Leela would beat GMs at non-blitz and GMs/IMs can mostly express in language why so and so move and so on.
I would agree on this point: I want chess solved before I die. If nobody else will do it then I will.

The results are very impressive, but it comes at a price: the NN, and hence the amount of knowledge it can store, has been made smaller. From the POV of solving chess, this is a slight negative.
This is a highly questionable conclusion IMHO. NNUE in chess is brand new and will unquestionably undergo many tweaks, changes, and experiments, whether in structure, size, or training methodology.

For the longest time, chess was dominated by the bean-counter approach, as CW is so fond of labeling it: the bare minimum of knowledge that cannot be overcome by sheer speed of search, resulting in Stockfish as its ultimate representative of late.

Then along came AlphaZero, which turned all this on its head, and in spite of a 900x deficit in nodes per second, was able to challenge this by pure quality of evaluation. It was a revolution of course.

This led to a rather curious battle of the extremes in computer chess: ultra knowledge and evaluation, running at incredibly low node counts, or ultra speed. There was no in-between... until now. True, the neural network sizes currently presented are a fraction of the size of the 256x20 (or larger) neural networks by DeepMind and then expanded by the Leela community, but these are still infinitely 'smarter' than the hand-crafted evaluation function honed over time. And they still enjoy colossal speeds.

Just as Leela was all the talk for a long time when it first began to flex its muscles, it is quite natural to see this happen with NNUE, and I am certain there will be plenty of developments to come. This is absolutely not the death of Leela, which would be both terrible and ridiculous, and is simply the natural next step trying to enjoy a 'best of both worlds'.
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."
corres
Posts: 3657
Joined: Wed Nov 18, 2015 11:41 am
Location: hungary

Re: The next revolution in computer chess?

Post by corres »

Ovyron wrote: Fri Jul 24, 2020 8:38 pm
corres wrote: Fri Jul 24, 2020 8:08 pm the chess power of default Stockfish(dev) and Stockfish NNUE is near the same
...
NNUE showcases the potential of Stockfish's eval improvement, as it has a cost and makes it search slower, but nothing stops Stockfish's eval from being improved and reach that same strength without slowdown. But then the new improved eval can be used for a new NNUE that is again slower but stronger, creating a virtuous cycle.
This is like a baby with 200 IQ and 200 EQ, tremendous potential, but still a baby. But she'll grow.
We only can hope for this kind of development.
User avatar
Ovyron
Posts: 4557
Joined: Tue Jul 03, 2007 4:30 am

Re: The next revolution in computer chess?

Post by Ovyron »

towforce wrote: Fri Jul 24, 2020 11:37 pm The results are very impressive, but it comes at a price: the NN, and hence the amount of knowledge it can store, has been made smaller. From the POV of solving chess, this is a slight negative.
No, it's that you don't need chess' solution, you just need an algorithm that gives you a perfect move for any chess position that you input. Then this algorithm can be very small.

And in fact, and I've mentioned this in another thread: pi contains this algorithm.

An exe is called a binary because it's just a series of 1s and 0s. The binary representation of pi contains all possible combinations of 1s and 0s, so not only it contains an algorithm that plays a perfect move for any chess position, it contains all of them, and it contains the shortest one.

And new approaches like NNUE, since they play better chess, approach perfection, and approach this algorithm. And if removing chess knowledge approaches it, knowledge isn't needed, and it's something positive.

The sad truth is that when the day comes you have a binary in your hands that plays perfect moves for any chess position, and its source code, you'll have no idea why it does it or how it works, all chess knowledge and its concepts are just human constructs.