Have you decided why Vas

[bob]

And that is an absolute fact that most users don't understand unfortunately or they don't want to,even worse,they consider themselves cheated by Vasik

Indeed so.

If there was a single, foolproof, universal method of calculating nodes per second, then every engine author had to embrace this method and incorporate it in their code.

I don't think people understand that if Vas wanted Rybka 3 to display 10 times the number of nodes calculated per second, he could just multiply the number with 10 and then (probably) we wouldn't have this discussion.

There's isn't an organization or 3rd-party jury who decides what the nodes per second number will be.

It's the engine author alone, and he chooses (yes chooses) the number he think fits the overall scheme of things.

So if Vas feels that the nodes per second for Rybka 3 is 5 times lower than Rybka 2, so be it. It makes no difference, what so ever.

So now... just forget about the nodes per second.

I do not have any idea where this is coming from. But to break it down.

1. "node" is _precisely_ defined in tree searching. Making a move leads to a new position, which is called a node. If you have a position with 20 legal moves, and you search exactly one ply (no extensions) deep.you search 20 nodes. no more, no less. One can argue about the "illegal moves" that are made, but if you "make illegal moves" then counting them is reasonable, and they represent a tiny fraction of the total nodes anyway and don't change the significance of that number.

2. "second" is a precisely defined unit of time. Standard everywhere

3. "per" as in a per b, means that for time interval B, there are "a" things happening.

So how is that _not_ precise? "depth" is far more vague, because of extensions and reductions and even pure forward-pruning. But nodes per second has been measured for 40 years and until this case came up, everyone was computing them the same way. Even in Belle, Ken connected an integrating timer to a pin that toggled each time a complete hardware cycle (make/select move/etc) was done, which was precisely NPS. Only exception I know of is Deep Blue and its predecessors, because they simply didn't take the time to count the nodes searched, and they estimated by using the theoretical max NPS per chip, and multiplied by the effective duty cycle (how often they were busy) to come up with a number that matched what everyone else considered NPS...

We have one program that is trying to hide things. That doesn't make it impossible for the rest of us to produce a real value for NPS that is comparable. Comparing NPS values reveals details. How long it takes to expand a node, which could be attributed to evaluation or complexities in the search brought on by pruning/etc.

[kgburcham]

Thanks Robert.
Also Robert thanks for being respectful in your replies in trying to help some of these guys understand. I noticed you kept cool and stayed with a technical answer and not personal.
You have posted this same info several times in the past, maybe they missed it.

I do not need kns, depth, total nodes, but it makes it more interesting, fun, etc...I look at my programs as soon to be outdated toys, a waste of money. Of course now programs have become trainers for the Super GM.

I am glad Vas decided not to eliminate the analysis info.

The Truth Will Set You Free
kburcham

[fern]

Else: It is not only how deep the search is and how many knods use to that, but who know what convoluted jumps that search does that gives Rybka such phenomenal strenght.
I am not programmer, I do not know much of this, but I suspect that Rybka search is not just the usual production of lines evaluated with tables and so on. And not only a bunch of knowledge to perform that evaluation. He discovered something else about how to perform the search.
As a vague example, remember how The King engine distributes full and selective search. The sheer fact it does full width here and selective there make a program totally different I believe.
It is my guess Vas use a sophisticated alternation between both full width and selective search.

who knows...
Fern

[bob]

Else: It is not only how deep the search is and how many knods use to that, but who know what convoluted jumps that search does that gives Rybka such phenomenal strenght.
I am not programmer, I do not know much of this, but I suspect that Rybka search is not just the usual production of lines evaluated with tables and so on. And not only a bunch of knowledge to perform that evaluation. He discovered something else about how to perform the search.
As a vague example, remember how The King engine distributes full and selective search. The sheer fact it does full width here and selective there make a program totally different I believe.
It is my guess Vas use a sophisticated alternation between both full width and selective search.

who knows...
Fern

remember that for a while, Vas was just a normal chess programmer. He participated in discussions here. He remembers the "mental disassembly" discussions about programs like chess genius, and fritz, and chessmaster, etc. we used to have many such discussions about programs, people devising test positions to detect what was going on, etc. And if you happen upon an idea that is unique, but not that hard to understand, then you could easily be concerned that with so many eyes looking on, the more information you provide, the easier it becomes to figure out the new idea. It is much easier to hide behind a curtain, and say "my NPS is much lower because I have so much knowledge in my program" and let that throw people off the track. Until someone actually disassembles your program and posts the details, including the fact that it is _not_ a knowledge-based evaluation-heavy program at all.

Programmers are a clever bunch. I used to play falcon 3.0 and got tired of having to have the CD installed for copy protection. And one day I fixed my version of falcon so that the CD was no longer required. Give a programmer enough motivation and they can discover details you thought were well-buried and hidden... Starve them for information only makes the process harder, but not completely impossible.

[Dr.Ex]

[/quote]
A chess program can give all sorts of insight into what is going on. For example, if you do a real 6-ply search, and display the _full_ PV, it is not that hard to look at the output for specific positions and discover details about the search. You have already seen lots of positions that break null-move, as one example. If you don't want me to look at a 6 ply search (so that I know the basic search depth) to see what is happening, then you make it impossible for me to produce a 6 ply search. If you don't want me to have any idea of how much of your program is search and how much is evaluation, then hide the NPS, and the nodes searched, and the depth, and don't even show the complete PV. Now it is _very_ hard to figure out what is going on because you can't obtain enough information to figure out much.

I can't imagine any other reason.[/quote]

Rybka is IMHO a brutal effective and fast searcher. It is as dumb as Toga for example. It has very little knowledge about chess.

[d] 1k1r3r/1ppqb1pp/p1n1pp1n/3pPb2/BP1P1P2/2P1BN2/P5PP/RN1Q1RK1 w - - 0 12

It thinks this position, which is easily won for white, is better for black.
Just like all other dumb engines like Toga etc. think it is better for black.
That says it all.

It is unfortunate that the game of Chess allows stupidity to prevail. We should just move on and search for another Board game where this can't happen.

[Jim Walker]

This discussion seems to come up with every new version of Rybka. Thank goodness Rybka is not coming out every 2-3 months now. What I don't understand is why it keeps coming up. If Vas wants to hide something about his program or not report the true nps of his program, why is it a concern to others? My opinion is that the true inner workings of a commercial program are nobodys business except the original creator. If you want to publish your source for the world to see it's your option but don't try to force your rules on others.
Jim

[Mike S.]

NPS = total nodes searched / total time used

it is just as simple as that...

If this is a de facto standard among chess programmers, it is still not completely clear to me:

(1) What about the positions at the end of the variations each, after the last moves? In other words, those where no more moves are being made, at a certain point during calculation time? By common sense, I think a node is something like a street junction where something branches off, still.

(2) Also, I have read that chess engines make pseudolegal moves, internally. Maybe that leads to a similar question for the node count, if the resulting positions of such moves are discarded (as I assume), and not evaluated or at least not searched deeper.

(3) Another possible distinction which would make sense to me is: Decide if hash table hits are counted or not. Because if the kN/s should serve as a performance figure, I am not sure if these should be included, or only those where the evaluation has to be calculated.

(4) Same for tablebase hits.

[Jim Walker]

A chess program can give all sorts of insight into what is going on. For example, if you do a real 6-ply search, and display the _full_ PV, it is not that hard to look at the output for specific positions and discover details about the search. You have already seen lots of positions that break null-move, as one example. If you don't want me to look at a 6 ply search (so that I know the basic search depth) to see what is happening, then you make it impossible for me to produce a 6 ply search. If you don't want me to have any idea of how much of your program is search and how much is evaluation, then hide the NPS, and the nodes searched, and the depth, and don't even show the complete PV. Now it is _very_ hard to figure out what is going on because you can't obtain enough information to figure out much.

I can't imagine any other reason.[/quote]

Rybka is IMHO a brutal effective and fast searcher. It is as dumb as Toga for example. It has very little knowledge about chess.

[d] 1k1r3r/1ppqb1pp/p1n1pp1n/3pPb2/BP1P1P2/2P1BN2/P5PP/RN1Q1RK1 w - - 0 12

It thinks this position, which is easily won for white, is better for black.
Just like all other dumb engines like Toga etc. think it is better for black.
That says it all.

It is unfortunate that the game of Chess allows stupidity to prevail. We should just move on and search for another Board game where this can't happen.[/quote]

You must be a great chess player. I'm not, so I would ask you to provide the winning line.
Jim

[Nimzovik]

DEEP SIGH ....... I prefer a relatively accurate node count if for no other reason than to know my computer is functioning or not. Some times I have to tweak the processes running to assure I have enuff resources available (allocated memory as windows does not always adjust this perfectly.) Now let me throw THIS in da mix.... I for one also loved to see the HASH meter on the older Fritz programs! Why continue analysing a position if the hash is used up?

[Titu]

Have we not already discussed this matter 100 times?

I believe Vas can show what he decides to.
- Vas is not running an academic project.
- Vas has not signed a UN treaty to show NPS in certain way.
- If Vas is trying to hide something so be it. He has already said that he does it in his own way. Vas is not following any other programs or oracles. Just look around how other commercial businesses run.
- If you don't like Vas's method don't buy his program.
- And if someone has not bought his program, it's seems strange that person sneers at Vas every chance he gets.

This dislike for Rybka from certain persons have been obvious from very beginning, almost showing signs of jealousy.

All of a sudden Vas comes to the computer chess and beats programs (20-0) that been around 30 years. And Vas does it in his own way. For some author this can be frustrating. Instead of showing any sort of acknowledgment for Vas, who has taken computer chess to a new level, lets pounce on the only thing that can be pounced on (compared to author's own program) every time, time and time again.