The real elo of lczero is in its name

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

Moderators: hgm, Rebel, chrisw

Daniel Shawul
Posts: 4185
Joined: Tue Mar 14, 2006 11:34 am
Location: Ethiopia

Re: The real elo of lczero is in its name

Post by Daniel Shawul »

I forgot you need to set "frac_alphabeta 0" to get scorpio-mcts-min.

The above setting is for scorpio-mcts-ab that is at the bottom of my list.
IQ
Posts: 162
Joined: Thu Dec 17, 2009 10:46 am

Re: The real elo of lczero is in its name

Post by IQ »

Daniel Shawul wrote: This is not coming from the test setup they used as some would claim but from this very glaring tactical weakness of MCTS that they claim is no problem. You would think Stockfish would get atleast one win from 100 games against alphazero. Why are they not playing AlphaZero on chess servers like they did for Go ?
I am always amazed that even knowledgeable people are not able to read the preprint properly. There were an additional 1200 games played, and Alpha Zero lost a couple of those. I agree it is not very informative to only show some nice wins, it would be even more interesting to have seen the losses.

I also agree that the rational given for using MCTS in the preprint paper is rather weak and unconvincing, stemming more from presenting a "unified" framework that "conquered" three games Go, Shogi and Chess.

I also believe that the NN of AZ with a more tradiditonal search would have been even stronger, but i do not believe the results are a hoax. Their Net (larger and more training) is probably much stronger than that of LCZ at this point in time and might catch more tactics without resorting to search. Also do not underestimate the ability of a very large net to learn openings - so while their results are not faked, we do not know yet how Alpha Zero would fare from more diverse openings than those used in the 1200 games match.

Also, you complain that Alpha Zero Chess does not play on the chess servers - while that would be nice, alpha zero go also (AFAIK) never played on any server. The last version of AG that played go on a server was the one called Master, predating the full zero approach.
User avatar
George Tsavdaris
Posts: 1627
Joined: Thu Mar 09, 2006 12:35 pm

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by George Tsavdaris »

Till now it is Scorpio - Leela +5 =1 -3
Surprisingly not a bad performance by Leela. It has a -79 ELO performance compared to Scorpio.
And it should have been +3 =3 -3, since Leela lost 2 completely drawn games because she didn't want to force an easy perpetual.


It reports something like 2000 to 3000 nodes per second on the 43 CPUs which is rather very bad as on a good GPU will get easy 1000.
So on a top GPU(Titan V or GTX 1080 Ti) would be much stronger. So perhaps to see its true strength we need very very good hardware so all our tests for ELO are biased. :D

You can watch the 20 game match here: http://tcec.chessdom.com/bonus/live.php
After his son's birth they've asked him:
"Is it a boy or girl?"
YES! He replied.....
User avatar
Nordlandia
Posts: 2821
Joined: Fri Sep 25, 2015 9:38 pm
Location: Sortland, Norway

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by Nordlandia »

George Tsavdaris: TCEC server need better graphics card for leela. If cpu handicap overall strength.
Nay Lin Tun
Posts: 708
Joined: Mon Jan 16, 2012 6:34 am

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by Nay Lin Tun »

Any way to download those pgn(archive)?
User avatar
George Tsavdaris
Posts: 1627
Joined: Thu Mar 09, 2006 12:35 pm

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by George Tsavdaris »

Nay Lin Tun wrote:Any way to download those pgn(archive)?
From there FILE->Save live PGN EVENT.
After his son's birth they've asked him:
"Is it a boy or girl?"
YES! He replied.....
User avatar
AdminX
Posts: 6340
Joined: Mon Mar 13, 2006 2:34 pm
Location: Acworth, GA

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by AdminX »

Nordlandia wrote:George Tsavdaris: TCEC server need better graphics card for leela. If cpu handicap overall strength.
The Twitch stream is not that bad. Plus you can Chromecast it to your TV or view it using a Fire Stick with Twitch installed.
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers
Hai
Posts: 598
Joined: Sun Aug 04, 2013 1:19 pm

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by Hai »

Nordlandia wrote:George Tsavdaris: TCEC server need better graphics card for leela. If cpu handicap overall strength.
TCEC should use best cpu(it had already) and best gpu(4x Titan V or at least 4x 1080ti).

It's not fair that Zero is handicapped by a factor of 10 or maybe more.
It would be also better if all engines must use gpu :lol: .
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by Milos »

Hai wrote:
Nordlandia wrote:George Tsavdaris: TCEC server need better graphics card for leela. If cpu handicap overall strength.
TCEC should use best cpu(it had already) and best gpu(4x Titan V or at least 4x 1080ti).

It's not fair that Zero is handicapped by a factor of 10 or maybe more.
It would be also better if all engines must use gpu :lol: .
L0 is not handicapped at all.
L0 on two 1080Ti in Sli mode is just marginally stronger than on one 1080Ti.
If you actually know how to make L0 run proportionally stronger on more GPUs please offer us some code.
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: Leela 43 cores Versus Scorpio 32 cores at TCEC

Post by Milos »

George Tsavdaris wrote:Till now it is Scorpio - Leela +5 =1 -3
Surprisingly not a bad performance by Leela. It has a -79 ELO performance compared to Scorpio.
And it should have been +3 =3 -3, since Leela lost 2 completely drawn games because she didn't want to force an easy perpetual.
It is already +13=3-4 after 20 games i.e. +160Elo for Scorpio (current weaker version of it anyway).
Funny how early statistics can be misleading. ;)