R4 vs Stockfish 1.9 x64

Not logged inRybka Chess Community Forum

Forum RybkaChess Help Search Register Login

Topic Rybka Support & Discussion / Rybka Discussion / R4 vs Stockfish 1.9 x64

1 2 3 4

By Uly

Date 2010-10-11 05:23

> I don't have any workarounds for SF1.9 going brain-dead when I move forward or backward.

My workarounds for such engines (Stockfish is highly critiqued for this just for being a top engine, in reality most engines are mediocre for score propagation) is to "emulate" persistent hash on, say, a text file, all persistent hash does is telling the engine what is a given move at a given depth, so one can emulate that by excluding the move (so that it's as if the engine knew the score of that move and was trying to find something better) to the given depth, and emulate higher depth by forcing the move. As long as one stores everything necessary, the time lost will mostly be about having to do it manually.

With Stockfish that was clunky since it basically starts from scratch at every action, so I'm very grateful that you recommended me to use it with MultiPV, I've been using MPV=4 and analyzing just as if I used exclude moves three times, but of course it is much faster, I think I lose 1.5-2 ply per move, but get x4 the information. It's the cases where I wouldn't have asked Stockfish for alternatives that are slower, but I just need to analyze with care. I also think that if the best move is the 5th one and Stockfish misses it, I shouldn't be analyzing those positions with Stockfish anyway.

It's knowing things like that that relieve me that Stockfish has such quirks, at least they're not as bad as Thinker's, whose workarounds were so time consuming that knowing them doesn't give any edge.

Vas probably went with that hash concept because it maximizes elo for games, that's why Naum has a separate parameter "PreserveHash" that works and makes it as good as Zappa for score propagation.

By TheHug

Date 2010-10-11 05:48

There is no doubt that this info was every useful. You would be surprised at how good Stockfish can find plans. The kicker is that its first and/or sometimes 2nd moves in one line are what you have to look at. Then I would use my other engines to help feel in the other parts of the plan. Its not to be overestimated that its not good enough to just find one or two good moves in a line. For it to work you usually have to put together a good plan or your advantage can fall apart. I think Vytron would know something about this. He is pretty good at micromanaging his advantage. It was very true in his 2nd game win vs ppiper.

By tomgdrums (****)

Date 2010-10-11 14:48

Do you think Stockfish is better in closed or open positions? I have never quite gotten a handle on Stockfish. And do you like the Stockfish book?

By Uly

Date 2010-10-12 01:10 Edited 2010-10-12 05:12

I haven't seen a dependence of the nature of the position, but I think Stockfish is better in complex ones, those that have many choices among both sides. What I've seen is better judgment.

Say, Rybka says move A, B, and C are 0.00, a player that only uses Rybka would be tempted to explore the variations and give them equal attention as they seem equally interesting. Now, in the same position Stockfish says move A is 0.00, move B is 0.16 and move C is 0.36, so an user using Stockfish can focus the resources on the third move, and Rybka will soon probably agree that C is not 0.00 but gives a slight edge to white, you've found the best move faster, and though you should also take a look at A and B, C should be analyzed deeper, Stockfish helps with finding candidate moves faster.

Someone could argue that with perfect play all these moves are actually drawn, so Rybka's score is more accurate, but that's not the point, if you want to win you've got to assume the opponent is going to blunder, and you have to "help" him by going to positions where it's easier to blunder, such as C.

I've been able to squeeze victories like that, eventually Stockfish would go bonkers and say the position is +1.00, while Rybka would stay calm with some 0.20, it turns out the position is very hard for black and Stockfish is right that is a desirable position to have as white (even if the 1.00 is very exaggerated.)

I've never taken a look at Stockfish's book.

By tomgdrums (****)

Date 2010-10-12 01:31

Thanks!!

I am going to start giving Stockfish a little more attention. I have always had trouble with it's up and down thing but your explanation helps me to figure it out a bit!

By Moz

Date 2010-10-12 22:17

> ...that's why Naum has a separate parameter "PreserveHash" that works and makes it as good as Zappa for score propagation.

Very true. Naum is a totally different beast with drastically reduced pruning and hash propagation (not to mention that it still handles the bishop duo better than the competition)