The Inside Story on Rybka

[Albert Silver]

Agreed. It is none of the user's business how the engine counts depth (i.e. which extensions and reductions it does) or selects its move. In fact this is proprietry information.

Maybe..but you should know what depth is being searched, KNPS, and a full PV.

Ply depths are relative. We know that engines are more or less selective, and even count plies differently. Not one single engine uses a non-reduced fullwidth ply-count, so all you can do is use the one given on a relative base. For example, If engine X provides decent results at 11 ply (its ply count) after 5 seconds, then you know you can probably get even better results after 25 seconds and 13 plies.

Try comparing ply depths between Junior, Hiarcs, and Fritz, even for short mates. A true exercise in futility IMO. I honestly only look at results when analyzing, and how long it usually takes to provide reliable ones.

Albert

[playjunior]

Indeed, on both counts. Some engines count QS nodes, others don't, some engines count hash cutoffs, some don't, some engines count only end leaves, others also count internal nodes, some engines count evaluations, and some of those do not evaluate in every node...

And some engines count a PV of 16 moves as depth=8, because half of the move are extensions, while others count the same PV as depth=20, because 4 of the moves were reduced, or part of QS...

Fine. Now tell me what is the justification of reporting nodecount/7 or whatever Rybka does.
And, as I pointed before, Chessbase explained how Junior's depth works.

In my city, there is a big screen in the center that reports temperature. I don't know where the measurements are done, how often they refresh, do they average through some previous measurements or just report the last number. Sometimes you feel it's a bit off, but it's quite fine most of the times.
Maybe in other cities there is a screen like that that reports temperature in a different way.
But if someone would divide the number to 7 to show how cool everything is, I would be unhappy. And the argument "there is no strict definition of temperature in the city" does not work.

[George Tsavdaris]

In my city, there is a big screen in the center that reports temperature. I don't know where the measurements are done, how often they refresh, do they average through some previous measurements or just report the last number. Sometimes you feel it's a bit off, but it's quite fine most of the times.
Maybe in other cities there is a screen like that that reports temperature in a different way.
But if someone would divide the number to 7 to show how cool everything is, I would be unhappy. And the argument "there is no strict definition of temperature in the city" does not work.

In my city we use °C for displaying temperatures.
If i go to a UK city i will see that they are displaying temperatures in a strange form: They take my temperature multiply it with 9/5 and add 32. Go figure! Should i feel unhappy?

BTW what kind of difference would make if Rybka reported 2000000 n/s instead of 300000 /ns but reported the same score and played the same move?

[playjunior]

In my city, there is a big screen in the center that reports temperature. I don't know where the measurements are done, how often they refresh, do they average through some previous measurements or just report the last number. Sometimes you feel it's a bit off, but it's quite fine most of the times.
Maybe in other cities there is a screen like that that reports temperature in a different way.
But if someone would divide the number to 7 to show how cool everything is, I would be unhappy. And the argument "there is no strict definition of temperature in the city" does not work.

In my city we use °C for displaying temperatures.
If i go to a UK city i will see that they are displaying temperatures in a strange form: They take my temperature multiply it with 9/5 and add 32. Go figure! Should i feel unhappy?

BTW what kind of difference would make if Rybka reported 2000000 n/s instead of 300000 /ns but reported the same score and played the same move?

I said why it should report real numbers. See y posts above: misleading everyone on purpose is not good. Especially when that everyone gives you his source code for free which is fruit of years of work, then answers your questions whatever they are. Then you fake the count, and all think in the wrong direction. Then someone disassembles you, and many think you deserved that.

[Terry McCracken]

Agreed. It is none of the user's business how the engine counts depth (i.e. which extensions and reductions it does) or selects its move. In fact this is proprietry information.

Maybe..but you should know what depth is being searched, KNPS, and a full PV.

Ply depths are relative. We know that engines are more or less selective, and even count plies differently. Not one single engine uses a non-reduced fullwidth ply-count, so all you can do is use the one given on a relative base. For example, If engine X provides decent results at 11 ply (its ply count) after 5 seconds, then you know you can probably get even better results after 25 seconds and 13 plies.

Try comparing ply depths between Junior, Hiarcs, and Fritz, even for short mates. A true exercise in futility IMO. I honestly only look at results when analyzing, and how long it usually takes to provide reliable ones.

Albert

I know all this. Chessbase/authors explain all this btw.

None of the other authors trie to hide anything and that's the problem we're dealing with.

Vas should show relative Ply Depth, Accurate Node Count and a Complete PV.

It's really that simple.

Terry

[rfadden]

I posted Vas's replies from the Rybka Forum, I gave them here in the other thread where I originally wrote the details...

Here's a repeat of my summary from that thread:

So I consider the Problem/Issue essentially resolved: The author has considered this input, and he indicates he's taking his own fresh approach in Rybka 3.0.

I like that! I feel that my concern has been heard and I'm happy with the overall process of discussion here and there.

Thanks everyone!
Rick

[George Tsavdaris]

I know all this. Chessbase/authors explain all this btw.

None of the other authors trie to hide anything and that's the problem we're dealing with.

How do you know? How do you know that Fritz for example shows its real nodes per second and not nps·7/8 +20 for example?

Vas should show relative Ply Depth, Accurate Node Count and a Complete PV.

It's really that simple.

No, it's more simple than this.
Vas should show what he wants in Rybka!
If someone does not like it, then he should not buy Rybka!

[rfadden]

The following might not be a popular thought but I did successfully add a patch to my copy of Rybka 1.0 beta that "fixed" one of the Obfuscations. As I have been reading this thread and thinking it occurs to me that I can just distribute a patch (not the original program) that each one of us can apply to our own binary and with this patch you will then see corrected Depth, Nodes, and NPS from the original Rybka binary as it runs.

So let me try to be clear. I have not completed the full set of patches that gives correct NPS for Rybka 1.0 beta, but I could do that with about a day's work. (Technical problem: I have to find a "hole" where I can put the additional code, and that is difficult but I believe it can be done).

So as the folks are saying above I would need to explain exactly what I am counting.

Also, people would naturally want to be careful about accepting and running patches. I know I would be a little wary about that. On the other hand I give my name, address and email with this site. I am not anonymous here so if there was a problem with my patch, I can be found, I'm right here (no bad intent here).

A further somewhat scary thought is that I could give a patch for each version of Rybka that adds on the same code for correct displays.

So this is just an idea. I'm not sure I want to give out any one of these patches because in later versions that would imply that I looked inside and I'm not necessarily going to do that.

So here I'm mentioning that I can develop a patch that gives the correct information for any version of Rybka and you can apply the patch to your own copy of the binary if you want. Then theoretically you could get Node Counts in any form that you wanted.

If I did develop a patch that you could use I think I would first send it to Vas and ask his opinion on the subject. Perhaps he would allow my patch to go out as a "Mod" or as a sanctioned mod. I could for example only proceed if he agrees that it's ok to do this.

This could be interesting, and if I did have Vas's permission then there would be no legality issues...

It's a thought...

[Terry McCracken]

I know all this. Chessbase/authors explain all this btw.

None of the other authors trie to hide anything and that's the problem we're dealing with.

How do you know? How do you know that Fritz for example shows its real nodes per second and not nps·7/8 +20 for example?

Vas should show relative Ply Depth, Accurate Node Count and a Complete PV.

It's really that simple.

No, it's more simple than this.
Vas should show what he wants in Rybka!
If someone does not like it, then he should not buy Rybka!

How do you know if your thermometer is working correctly? This is a silly course of reasoning. They most likey didn't hide their work. A programmer would know and it would have been revealed by now.
If I'm wrong, well then I'm sure people will investigate the matter after this.

Terry

[GenoM]

If someone does not like it, then he should not buy Rybka!

Of course, if the buyer knows exactly what he's buying. But the case is different: the buyer doesn't know this because the only way he can know this is when a seller explains him what exactly he's buying.
Simple, isn't it?