Lc0 Vs Stockfish Match.

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.
mwyoung
Posts: 2317
Joined: Wed May 12, 2010 8:00 pm

Lc0 Vs Stockfish Match.

Post by mwyoung » Fri Feb 22, 2019 2:33 am

I would be shocked if Lc0 wins the current super final. In my testing Stockfish is stronger then the 3xxxx NN of Lc0.

I will be equal as shock if Lc0 does not win it all in the next season. I have been charting the progress of the 5xxxx NN of Lc0. And its growth in strength is impressive. Just in the past 24 hours the 5xxxx NN have increased 30 elo against Stockfish in my testing. If that trend continues and does not flat line. It is only a matter of a couple of weeks before Stockfish is clearly overthrown.
Professing themselves to be wise, they became fools,
take on me. Foes 0.

Dann Corbit
Posts: 11495
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Lc0 Vs Stockfish Match.

Post by Dann Corbit » Fri Feb 22, 2019 2:39 am

mwyoung wrote:
Fri Feb 22, 2019 2:33 am
I would be shocked if Lc0 wins the current super final. In my testing Stockfish is stronger then the 3xxxx NN of Lc0.

I will be equal as shock if Lc0 does not win it all in the next season. I have been charting the progress of the 5xxxx NN of Lc0. And its growth in strength is impressive. Just in the past 24 hours the 5xxxx NN have increased 30 elo against Stockfish in my testing. If that trend continues and does not flat line. It is only a matter of a couple of weeks before Stockfish is clearly overthrown.
Stockfish has also been making big gains in the past weeks.

I expect that a hybrid strategy will pass them both within a few months.

There are already preliminary versions.
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.

mwyoung
Posts: 2317
Joined: Wed May 12, 2010 8:00 pm

Re: Lc0 Vs Stockfish Match.

Post by mwyoung » Fri Feb 22, 2019 3:34 am

Dann Corbit wrote:
Fri Feb 22, 2019 2:39 am
mwyoung wrote:
Fri Feb 22, 2019 2:33 am
I would be shocked if Lc0 wins the current super final. In my testing Stockfish is stronger then the 3xxxx NN of Lc0.

I will be equal as shock if Lc0 does not win it all in the next season. I have been charting the progress of the 5xxxx NN of Lc0. And its growth in strength is impressive. Just in the past 24 hours the 5xxxx NN have increased 30 elo against Stockfish in my testing. If that trend continues and does not flat line. It is only a matter of a couple of weeks before Stockfish is clearly overthrown.
Stockfish has also been making big gains in the past weeks.

I expect that a hybrid strategy will pass them both within a few months.

There are already preliminary versions.
I have not seen any major gains with Stockfish in the past few weeks. And in the past 2 weeks since the 8th they have released some new versions starting today. And I have starting testing with the new versions of Stockfish. But going by the Stockfish data. I don't expect major improvement, but just a incremental improvement. But the testing will reveal all. And as always I will post or stream the results.

And a hybrid is fine, but Lc0 is plugging some major tactical holes with the new version. And if this trend continues a hybrid maybe a moot point.
Stockfish.png
Stockfish.png (174.87 KiB) Viewed 2277 times
Professing themselves to be wise, they became fools,
take on me. Foes 0.

User avatar
M ANSARI
Posts: 3466
Joined: Thu Mar 16, 2006 6:10 pm

Re: Lc0 Vs Stockfish Match.

Post by M ANSARI » Fri Feb 22, 2019 6:29 am

Yes it would be pretty surprising if Lc0 could win TCEC 14 as Lc0 has just way too many holes and weaknesses to patch up, and SF is very capable and seasoned veteran. But I would be even more surprised if Lc0 doesn't win TCEC 15 as it seems that the weaknesses there are should be patched up pretty quickly. Almost every new engine that comes up goes through this bug swatting phase ... it is just that Lc0 ... even with all these weaknesses, is still crazy strong and is able to compete. I do think that for the short term, most likely an engine that uses NN for positional play and AB for tactical and endgame play would be unbeatable. But that is just a band aid patch as it probes only the immediate move to be played. I think the holy grail is to have an NN engine that can play tactics and endgames just like the top AB engine, but have that power to use these strong tactics deep in its search.

Leo
Posts: 957
Joined: Fri Sep 16, 2016 4:55 pm
Location: USA/Minnesota
Full name: Leo Anger

Re: Lc0 Vs Stockfish Match.

Post by Leo » Fri Feb 22, 2019 3:24 pm

What is the current score in the TCEC 14? I thought LCO was dominating based on comments from talkchess.
Advanced Micro Devices fan.

Werewolf
Posts: 1304
Joined: Thu Sep 18, 2008 8:24 pm

Re: Lc0 Vs Stockfish Match.

Post by Werewolf » Fri Feb 22, 2019 3:28 pm

mwyoung wrote:
Fri Feb 22, 2019 2:33 am
I would be shocked if Lc0 wins the current super final. In my testing Stockfish is stronger then the 3xxxx NN of Lc0.

I will be equal as shock if Lc0 does not win it all in the next season. I have been charting the progress of the 5xxxx NN of Lc0. And its growth in strength is impressive. Just in the past 24 hours the 5xxxx NN have increased 30 elo against Stockfish in my testing. If that trend continues and does not flat line. It is only a matter of a couple of weeks before Stockfish is clearly overthrown.
I find it striking that Test 50 was not designed to be this monster you report, more an experiment of ideas. It is Test 60 which is hoped to finally replace Test 40 (itself a new version with the latest findings from Alpha Zero).

So - - your news is very exciting indeed. If Test 50 is so strong where will we go in the future? But for now neither T50 nor T40 are complete, and it's not clear which one will end up being better.

User avatar
M ANSARI
Posts: 3466
Joined: Thu Mar 16, 2006 6:10 pm

Re: Lc0 Vs Stockfish Match.

Post by M ANSARI » Fri Feb 22, 2019 5:50 pm

I think for Lc0 there needs to be a way to figure out how to use the available hardware for tactical calculations. A lot of new stuff coming out with the new GPU's and I am not sure it is being utilized as best it could in either hardware efficiency or software coding. I think self learning can only take you so far but you need good tactical ability and proper endgame play, which is only possible through accurate calculation. Ironically the tactical weakness is also present with human play ... but here you have plenty of hardware to circumvent that weakness.

User avatar
Graham Banks
Posts: 34356
Joined: Sun Feb 26, 2006 9:52 am
Location: Auckland, NZ

Re: Lc0 Vs Stockfish Match.

Post by Graham Banks » Fri Feb 22, 2019 8:45 pm

Leo wrote:
Fri Feb 22, 2019 3:24 pm
What is the current score in the TCEC 14? I thought LCO was dominating based on comments from talkchess.
After 94 games out of 100, SF leads by a point.
gbanksnz at gmail.com

Leo
Posts: 957
Joined: Fri Sep 16, 2016 4:55 pm
Location: USA/Minnesota
Full name: Leo Anger

Re: Lc0 Vs Stockfish Match.

Post by Leo » Fri Feb 22, 2019 8:59 pm

Graham Banks wrote:
Fri Feb 22, 2019 8:45 pm
Leo wrote:
Fri Feb 22, 2019 3:24 pm
What is the current score in the TCEC 14? I thought LCO was dominating based on comments from talkchess.
After 94 games out of 100, SF leads by a point.
Thanks. I just saw that. What a battle.
Advanced Micro Devices fan.

User avatar
MikeB
Posts: 4354
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Lc0 Vs Stockfish Match.

Post by MikeB » Fri Feb 22, 2019 9:13 pm

Graham Banks wrote:
Fri Feb 22, 2019 8:45 pm
Leo wrote:
Fri Feb 22, 2019 3:24 pm
What is the current score in the TCEC 14? I thought LCO was dominating based on comments from talkchess.
After 94 games out of 100, SF leads by a point.
That qualifies as "epic". 😊
Image

Post Reply