Three fishes vs Leela

[Laskos]

I do not understand the name leelafish.

https://github.com/killerducky/lc0/wiki/LeelaFish

LeelaFish is a modified version of the Lc0 engine that gets help from an outside (auxiliary) engine.

There is Lc0 and another engine and the second engine does not have to be stockfish so why the name leelafish?

So, in case of a Fish as aux-engine it should be called Fishleela, shouldn't it?

My trials with combinations of Fish and Leela show much better effect at supporting SF with Leela instead of other way round, because SF is much better at hash-learning, so if moves differ, SF is to be brought to change its mind much better as for getting points of moves and lines given in from outside of own pondering. One could argue, that might be because of moves from Leela being better more often, but that doesn't seem to be so to me as for evaluating moves along longer lines at Forward- Backward. To say it shortly, backward analysis with SF is much more effective than with Leela, if you compare time to solution and time to eval.

To talk about it relevantly, we would have to discuss certain positions always only of course.

You see that best at interactive analysis with using both engines, which of course isn't to be measured Elo- wise except at corr.-chess, but as for Nucleus I do prefer SF as master engine too. (LC0 wouldn't have to give first ply- suggestions at all anyhow as master engine in Nucleus- combi.)

It might depend on the hardware used. If I use on one hand Leela on RTX 2070 GPU + 2 cores and on the other SF on 2 cores (I have 4 fast cores CPU), Leela is surely to be preferred as the main engine. Not only it is very much stronger, but aside tactical blunders in complicated positions and some endgames, Leela's moves are almost universally better than SF ones. It seems you want to analyze already played games forward and backward using the regular hash scheme of SF, and you even have some "time to solution", are you checking for blunders? In that case it's better to use SF without any Leela. Strength-wise and say correspondence and centaur chess, in my conditions, Leela is clearly the main engine having the say in close to 90% of the moves. LeelaFish still seems to not be able to correct some endgames (just an impression) during the gameplay (with Leela being the main engine).

[peter]

...
Leela is surely to be preferred as the main engine. Not only it is very much stronger, but aside tactical blunders in complicated positions and some endgames, Leela's moves are almost universally better than SF ones. It seems you want to analyze already played games forward and backward using the regular hash scheme of SF, and you even have some "time to solution", are you checking for blunders? In that case it's better to use SF without any Leela. Strength-wise and say correspondence and centaur chess, in my conditions, Leela is clearly the main engine having the say in close to 90% of the moves. LeelaFish still seems to not be able to correct some endgames (just an impression) during the gameplay (with Leela being the main engine).

You say it, Kai,
"aside tactical blunders and some endgames"

And aside positions, tactical or positional, where Leela doesn't get the clue from herself in a reasonable time, not even of matter, if SF does or doesn''t do better in the position at stand-alone pondering only.

If there's any such position, where you have to go into the lines of solution, 1, 2, 3 plies deep only or even 10 moves and more, if SF has "got" it with output and eval at the "points" along the line, you can keep the later on to be "found" moves and correct evals in hash (seeing them in output) going backward the lines, with Leela you can't in a comparable time.

It loses key moves and evals after much less plies backward than SF and or takes much more time to "refind" them.

That's what I call hash-learning, and that's something SF (and other good A-B-engines) to me and my experience (and that's a thing I do have much experience with in using engines e.g. for corchess, it's my way of interactive analysis, by the way it's the way of choice of most chess players I know, using engines as analysis- tools at all, not only letting play engines against engines all the time) do much better then NN- engines still, "Leelafish" (which should be called Fishleela being an Leela rather than a fish) is about the same as Leela itself in this respect as an analysis- engine.

Leela is not to be "taught" moves better than its own ones, not by a Fish (so I'd take it rather as slave than as master in Nucleus too) and not by man.

Leela takes much more moves to be "shown" tactial or positional "points" of deeper lines, if she doesn't "get" them on her own, but much more than this, SF is much faster in backwarding those lines without losing moves and eval out of output again.

Of course it depends on hardware too, but I'll be back in a while with the one or the other one position to show to you practically, what I mean.
Just need a little more time waiting for the answer of a corr.-chess friend of mine by mail, if I may take a recently played game of his, which would be an especially good example to me.

And I don't want to make that posting too long.
I think you know what I mean anyhow already, you just don't believe it, so let's discuss it, if you really like too, with some example- postions.

[gonzochess75]

So it is not really Brainfish but Stockfish.

Not good. I was fired up that Brianfish would see some action.

There is information in this as well. And if you are really interested in knowing the relative strength of these four engines on particular hw at particular tc with particular params, this tournament is not the way you'd get that info.