jp wrote: ↑Fri Jan 31, 2020 10:59 pm
I've only been posting Komodo's first choice because it's at such low depths (27-29) that it should not influence anyone (and if I wait for a move to be played, I will have forgotten what Komodo's choice was).
zullil wrote: ↑Fri Jan 31, 2020 10:27 pm
I rarely use Komodo, but I do like the "stability" of its evaluations, compared to Stockfish. Of course, stability of bad evaluations won't help.
Does Komodo have a default contempt setting? (Some of the evals I posted have been with White to move, others with Black to move.)
zullil wrote: ↑Fri Jan 31, 2020 10:27 pm
Also wondering what Stockfish thought about 13...h5. Top choice? Blunder?
SF10 at depth 29 gives first choice 13...h6 (-1.45).
Well, 13...h6 wasn't in Komodo's top three either.
zullil wrote: ↑Sat Feb 01, 2020 12:10 am
Well, 13...h6 wasn't in Komodo's top three either.
Really? Oh dear, it'd save time if you just told us what were those secret top 3
Sure. Just wanted to make sure that doing so wouldn't be considered interference. According to Komodo-13.2.5, for the position after White's 13th move:
13...Ne7 (-1.15)
13...O-O (-1.07)
13...Nf6 (-0.94)
This was using MultiPV = 3 at depth 48 after about 30 hours of search.
By the way, Komodo's opinion of 13...h5 is growing. The eval of the current position, after 13...h5, is now at -1.04 at depth 45.
Ethereal 11.89 at depth 49 and 182 billion nodes gives -0.98 (previous iteration was at -1.09, it's never fully stable) with b2-b3 as best move and Ng5 as second best. I'm giving the moves, because what Ethereal thinks should really not influence mmt's choice, and these two are already well-known candidate moves in this position.
Ovyron wrote: ↑Sat Feb 01, 2020 12:30 am
While we wait, I did another chessmap zoom into the Grob on this area
But what does a chessmap really tell us? Is it just relative sizes of variations in one particular database, which would only tell us about that database, not about the opening itself?
ADHD ...just joking, forgive me ...im seriously curious though what mmt will hatch here, i could imagine that LC still has a good feeling about best moves here but this assumption comes out of the sky because i cant use LC due to nonexistent GPU, interesting game, thx to both players !
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.
Spliffjiffer wrote: ↑Sat Feb 01, 2020 4:43 pm
i could imagine that LC still has a good feeling about best moves here but this assumption comes out of the sky because i cant use LC due to nonexistent GPU
We've seen before that after a long enough time Leela and SF often start to go for the same moves, sometimes for many plies, even if we doubt they understand the position and think they may be playing it "wrongly".
Thanks Zullil, I'll confess that I didn't check 13...Ne7 at all, so I can't comment on its quality vs. 13...h5. But in several variations white equalizes by playing h5 themselves first, and in others black wins by playing that knight to f4 (they even were very worrying, as apparently white can force a lot of exchanges, but they seem fine) which can't be done if white kicks it to e7.
What would be funny is if my move was inconsequential as there's many ways to skin a cat
[d]3qk2r/p2n1pp1/1rpbb1n1/3p3p/3P3P/1PNQ1N2/P3PPB1/R1B1K2R b KQk - 0 14
This was really hard to choose. LC0 by itself prefers 14. Ng5 (-1.27) but when pushed further along the Ng5 line it weakens to <-1.40. When pushed ahead with 14. b3, it gets the exact same score -1.27. Many moves are semi-forced with Ng5. SF 11 sees trouble ahead for both of these options with LC0's preferred responses (around -1.50 to -1.65). Now I'm quite sure that SF's eval is better in this position because of how deep it can see.
[d]3qk2r/p2n1pp1/1rpbb1n1/3p3p/3P3P/1PNQ1N2/P3PPB1/R1B1K2R b KQk - 0 14
This was really hard to choose. LC0 by itself prefers 14. Ng5 (-1.27) but when pushed further along the Ng5 line it weakens to <-1.40. When pushed ahead with 14. b3, it gets the exact same score -1.27. Many moves are semi-forced with Ng5. SF 11 sees trouble ahead for both of these options with LC0's preferred responses (around -1.50 to -1.65). Now I'm quite sure that SF's eval is better in this position because of how deep it can see.
Komodo at depth 47 with MultiPV = 3 had
14. b3 (-1.06)
14. Ng5 (-1.10)
14. Bh3 (-1.32)
Interestingly, Komodo had Ng5 as its first choice until depth 46.