LCZero: Progress and Scaling. Relation to CCRL Elo

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Albert Silver
Posts: 2869
Joined: Wed Mar 08, 2006 8:57 pm
Location: Rio de Janeiro, Brazil

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by Albert Silver » Tue May 15, 2018 8:09 pm

yanquis1972 wrote:
Tue May 15, 2018 7:03 pm
did want to add that just after i posted that, i watched a game against rybka 3 that quickly boiled down to an early endgame. rybka evaluated drawish & stayed there, leela +2 or so (ended drawn). i realized the answer there is pretty obvious too; not every game contains an endgame, but most would've, & the large majority probably ended decisively.

also forgot to mention the other hardware aspect to the tactical problem; while we're waiting for millions of games & hoping she stumbles upon the solution often enough to learn it, i'm guessing google used training h/w that could calculate several orders beyond what lc0 does. but i'm hopefully wrong & it was volume-focused.

stumbled on this graph (re strength vs stockfish based on time per move) which is interesting as well
Image
If you are really interested in the project, and how it worked, you should not 'guess' what Google did, or 'stumble' on the graph they published and just read their paper and the countless articles on it.
"Tactics are the bricks and sticks that make up a game, but positional play is the architectural blueprint."

jp
Posts: 840
Joined: Mon Apr 23, 2018 5:54 am

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by jp » Tue May 15, 2018 8:26 pm

Good question, though, about generated LC0 training games vs generated AZ training games. (Time/nps per move comparison?)

yanquis1972
Posts: 1762
Joined: Tue Jun 02, 2009 10:14 pm

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by yanquis1972 » Tue May 15, 2018 9:20 pm

Albert Silver wrote:
Tue May 15, 2018 8:09 pm
yanquis1972 wrote:
Tue May 15, 2018 7:03 pm
did want to add that just after i posted that, i watched a game against rybka 3 that quickly boiled down to an early endgame. rybka evaluated drawish & stayed there, leela +2 or so (ended drawn). i realized the answer there is pretty obvious too; not every game contains an endgame, but most would've, & the large majority probably ended decisively.

also forgot to mention the other hardware aspect to the tactical problem; while we're waiting for millions of games & hoping she stumbles upon the solution often enough to learn it, i'm guessing google used training h/w that could calculate several orders beyond what lc0 does. but i'm hopefully wrong & it was volume-focused.

stumbled on this graph (re strength vs stockfish based on time per move) which is interesting as well
Image
If you are really interested in the project, and how it worked, you should not 'guess' what Google did, or 'stumble' on the graph they published and just read their paper and the countless articles on it.
then i'm not 'really' interested.

duncan
Posts: 10525
Joined: Mon Jul 07, 2008 8:50 pm

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by duncan » Tue May 15, 2018 9:56 pm

Laskos wrote:
I just posted in this thread the result for ID292, it is the strongest ever (the standard v0.10 CPU and GPU build on master).
so this nightmare may soon be over.

(although perhaps I am taking this too seriously)

Milos
Posts: 3387
Joined: Wed Nov 25, 2009 12:47 am

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by Milos » Wed May 16, 2018 12:32 am

Albert Silver wrote:
Tue May 15, 2018 8:09 pm
If you are really interested in the project, and how it worked, you should not 'guess' what Google did, or 'stumble' on the graph they published and just read their paper and the countless articles on it.
He would also know that particular Figure 2 is totally bogus at least concerning SF's performance.

jp
Posts: 840
Joined: Mon Apr 23, 2018 5:54 am

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by jp » Wed May 16, 2018 7:17 am

Milos wrote:
Wed May 16, 2018 12:32 am
He would also know that particular Figure 2 is totally bogus at least concerning SF's performance.
The blue line performance in that Figure 2 is terrible anyway, even if they provide green lines for distraction.

yanquis1972
Posts: 1762
Joined: Tue Jun 02, 2009 10:14 pm

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by yanquis1972 » Wed May 16, 2018 9:24 am

do they say "this graph is bogus btw" in their own paper, or would i have to read the dozens of articles to find that?

i guess the main problem is, while chess & chess engines fascinate me & are an on/off hobby, i have no knowledge in the sciences. did i miss a forum rule? is there one particular degree i should pursue or will i need a couple? is post-grad at least recommended? would appreciate any further advice on what i need to do before posting on the internet.

jp
Posts: 840
Joined: Mon Apr 23, 2018 5:54 am

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by jp » Wed May 16, 2018 9:51 am

yanquis1972 wrote:
Wed May 16, 2018 9:24 am
do they say "this graph is bogus btw" in their own paper, or would i have to read the dozens of articles to find that?
That would make the paper worth reading. 8-)

mirek
Posts: 42
Joined: Sat Mar 24, 2018 3:18 pm

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by mirek » Wed May 16, 2018 10:13 am

Milos wrote:
Wed May 16, 2018 12:32 am
Albert Silver wrote:
Tue May 15, 2018 8:09 pm
If you are really interested in the project, and how it worked, you should not 'guess' what Google did, or 'stumble' on the graph they published and just read their paper and the countless articles on it.
He would also know that particular Figure 2 is totally bogus at least concerning SF's performance.
Then I have a question that I already asked before. Can you show measured numbers of true SF8 scaling on 64 cores for i.e. 1s / move vs 1min / move? Something other than your word that would prove your claim.

JJJ
Posts: 1287
Joined: Sat Apr 19, 2014 11:47 am

Re: LCZero: Progress and Scaling. Relation to CCRL Elo

Post by JJJ » Wed May 16, 2018 10:34 am

based on selfplay testing, I see progress are coming back and fast now ! I can raise my hope again.

Post Reply