Question to Larry Kaufman about Rybka

[metax]

You said about the 3 plies that Rybka subtracts from the real depth that this is done "on the ground that they are highly selective". First of all, most other top programs are also highly selective on the first few plies but do report them anyway. Secondly, it is known that Rybka shows only a small fraction of the real nodes searched and nps. I think here you can't find any explanation like "the nps not displayed are highly unstable". So I don't see any technical reason to report wrong depths and node counts.

I rather believe that this is a marketing trick to keep up the myth of an 'intelligent' Rybka which is stronger than all other engines even though searching less positions and looking ahead less plies among people who are rather unfamiliar with chess programing.

Could you please comment on this?

[alpha123]

From what I've heard, Vas just likes smaller numbers . Also, it's supposed to be more of a benchmark sort of thing (the NPS).

Peter

[zamar]

Because Rybka has been clear number 1 in all rating list for quite a long time, I think there is no need for "cheap marketing tricks".

It might simply be a question of lazyness.

[metax]

Because Rybka has been clear number 1 in all rating list for quite a long time, I think there is no need for "cheap marketing tricks".

It might simply be a question of lazyness.

I always divide all my reported UCI numbers by 15 (or so) out of lazyness, too.

[zamar]

If you take a look at ippolit, you'll find function named "slide_search". I believe that those nodes were not counted in (for lazyness or for any other reason).

[mjlef]

If you take a look at ippolit, you'll find function named "slide_search". I believe that those nodes were not counted in (for lazyness or for any other reason).

It seems to count moves made by white, black and null moves and adds them up. So moves generated and pruned without being played on the board do not get counted. Since the routines used to make moves are used at all nodes, then the count is accurate. I think many other programs also just count moves actually made on the baord. It is interesting it also counts null moves. I have not been doing that.

[lkaufman]

I don't know the original motivation for reducing the reported depth by 3, as it happened before I joined Rybka. However there was a reasonable justification for this, because those last three plies were pruned much more heavily than in other programs prior to Rybka 1. Even now, it is apparent that Rybka and her derivatives must be pruning more than other programs, because Robbolito (which reports proper depths) outsearches all unrelated programs by a ply or more in general. How this is achieved without a loss in strength at any given ply depth is the secret of Rybka's strength, at least it's a secret to me despite my long work on Rybka's evaluation. The node count is another matter, I can't justify the low Rybka counting method.

[Gian-Carlo Pascutto]

because those last three plies were pruned much more heavily than in other programs prior to Rybka 1.

I complained about this before, and I'll do so again: you have no way of knowing how Shredder, Junior, Fritz, Hiarcs and Sjeng work unless you've broken into our offices, so please don't claim you do, or I'll call the cops

Even now, it is apparent that Rybka and her derivatives must be pruning more than other programs, because Robbolito (which reports proper depths) outsearches all unrelated programs by a ply or more in general.

You can achieve the same by making the program twice faster in NPS without any difference to pruning (and incidentally, RobboLito is much faster than most of those programs), so your conclusion doesn't follow from the data at all.

[mcostalba]

You can achieve the same by making the program twice faster in NPS without any difference to pruning (and incidentally, RobboLito is much faster than most of those programs), so your conclusion doesn't follow from the data at all.

Twice faster is impossible IMHO. I have spent long hours at profiler to squeeze even the last drop out of SF and I know the pure speed increase we had through the versions starting from 1.3 is on average much smaller and from 1.4 to 1.5.1 when I've done a super optimization hard work I was able to get perhaps up to 8-10% of more speed, no more.

Today, if someone could squeeze another 5% out of SF (pure speed, no other tricks) is a profiling genius

[Gian-Carlo Pascutto]

Yes, but Stockfish is about as fast as RobboLito, last time I looked, so that's not surprising.