Similarity tester - 2nd generation - BETA

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.
User avatar
Laskos
Posts: 9545
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: Similarity tester - 2nd generation - BETA

Post by Laskos » Mon Aug 26, 2019 5:13 pm

Branko Radovanovic wrote:
Mon Aug 26, 2019 4:56 pm
Rebel wrote:
Mon Aug 26, 2019 8:53 am
Did some Lc0 testing - http://rebel13.nl/html/kai.html

SF vs Lc0 very low similarity.
Indeed, and that's not surprising. No real difference between SF8 and SF10 vs NNs - that seems to prove my impression was not correct, though. And, finally, of all AB engines SF is the most similar to Lc0 (while remaining very far nevertheless) - that would again make sense to me.
Check this recent thread:
viewtopic.php?f=7&t=71610

Branko Radovanovic
Posts: 59
Joined: Sat Sep 13, 2014 2:12 pm

Re: Similarity tester - 2nd generation - BETA

Post by Branko Radovanovic » Tue Aug 27, 2019 9:34 am

chrisw wrote:
Mon Aug 26, 2019 6:52 am
Branko Radovanovic wrote:
Mon Aug 26, 2019 12:11 am
My impression - not sure if that's true or not, so I'd like to see it tested - is that SF10's play is more similar to LC0 than e.g. SF9 was. That would make sense because if LC0's style is the "chess truth", devoid of preconceptions, one would expect AB engines to gradually approach it.
not necessarily, if chess is a draw, or basically a stable game, as people claim. There’s many ways to get to a draw.
A very good point. If chess is a draw, there are indeed many ways to get it with perfect play, playing against a perfect player, and all are equally good.

However, things get more complicated for non-perfect play. While, with perfect play, 1.d4 might be a draw, and - for all we know - 1.Na3 might be a draw too, I'd argue that for a non-perfect player 1.d4 is a much better choice than 1.Na3. So, while perfect players might draw each other with possibly wildly different styles of play, the reality is that "real-world-good" moves should be preferred over "not-so-real-world-good" moves, which significantly constrains one's choices. But in the end I have to agree: whether these practical limitations will eventually produce One True Playing Style or not remains to be seen.

chrisw
Posts: 2209
Joined: Tue Apr 03, 2012 2:28 pm

Re: Similarity tester - 2nd generation - BETA

Post by chrisw » Tue Aug 27, 2019 6:40 pm

Branko Radovanovic wrote:
Tue Aug 27, 2019 9:34 am
chrisw wrote:
Mon Aug 26, 2019 6:52 am
Branko Radovanovic wrote:
Mon Aug 26, 2019 12:11 am
My impression - not sure if that's true or not, so I'd like to see it tested - is that SF10's play is more similar to LC0 than e.g. SF9 was. That would make sense because if LC0's style is the "chess truth", devoid of preconceptions, one would expect AB engines to gradually approach it.
not necessarily, if chess is a draw, or basically a stable game, as people claim. There’s many ways to get to a draw.
A very good point. If chess is a draw, there are indeed many ways to get it with perfect play, playing against a perfect player, and all are equally good.

However, things get more complicated for non-perfect play. While, with perfect play, 1.d4 might be a draw, and - for all we know - 1.Na3 might be a draw too, I'd argue that for a non-perfect player 1.d4 is a much better choice than 1.Na3. So, while perfect players might draw each other with possibly wildly different styles of play, the reality is that "real-world-good" moves should be preferred over "not-so-real-world-good" moves, which significantly constrains one's choices. But in the end I have to agree: whether these practical limitations will eventually produce One True Playing Style or not remains to be seen.
LC0 chess can never be “truth” because it is a statistical machine which can never output e4 with p = 1.0
MCTS is statistical. The NN outputs are statistical.

jp
Posts: 841
Joined: Mon Apr 23, 2018 5:54 am

Re: Similarity tester - 2nd generation - BETA

Post by jp » Thu Aug 29, 2019 11:10 am

chrisw wrote:
Tue Aug 27, 2019 6:40 pm
LC0 chess can never be “truth” because it is a statistical machine which can never output e4 with p = 1.0
MCTS is statistical. The NN outputs are statistical.
And in the present it's even odder to make any suggestions about truth, when Lc0 is losing games, as are all other engines.

Post Reply