Rybka node counting issue: an easy solution!

[Jouni]

Multiply by ten (10x). I compared Rybka 1.0 and Strelka 2.0 and average
ratio in node speed was 11,6. So 10 is good estimate in whole game.

Jouni

[kgburcham]

I agree Jouni. I posted the other day to just add a zero to the kns in Rybka eval window. Doesnt matter to me anyway. I will just observe how all of this kns plays out.

kburcham

[Dann Corbit]

Multiply by ten (10x). I compared Rybka 1.0 and Strelka 2.0 and average
ratio in node speed was 11,6. So 10 is good estimate in whole game.

Jouni

Why does anyone think that the node count matters at all anyway?

The proof of the pudding is in the eating. Rybka plays strong, that's enough.

Junior counts plies differently than other programs (for example). Simply discount leaf nodes and qnodes and suddenly our node count vanishes to a small number. Perhaps Vas defines a node as 'a fully searched node'.

At any rate, the node count is not actually used for anything except FYI. I really don't understand why people are upset about it.

[bob]

Multiply by ten (10x). I compared Rybka 1.0 and Strelka 2.0 and average
ratio in node speed was 11,6. So 10 is good estimate in whole game.

Jouni

Why does anyone think that the node count matters at all anyway?

The proof of the pudding is in the eating. Rybka plays strong, that's enough.

Junior counts plies differently than other programs (for example). Simply discount leaf nodes and qnodes and suddenly our node count vanishes to a small number. Perhaps Vas defines a node as 'a fully searched node'.

At any rate, the node count is not actually used for anything except FYI. I really don't understand why people are upset about it.

I can think of just one valid reason: Some are interested in comparing programs, for any of several reasons. And one feature that defines a program's strength is "knowledge". For years everyone has compared their speed to that of Fritz, or genius, or any of a dozen other programs, and used that to draw conclusions. It is no longer possible, since the data is obfuscated to mislead such investigations.

[Terry McCracken]

Multiply by ten (10x). I compared Rybka 1.0 and Strelka 2.0 and average
ratio in node speed was 11,6. So 10 is good estimate in whole game.

Jouni

Why does anyone think that the node count matters at all anyway?

The proof of the pudding is in the eating. Rybka plays strong, that's enough.

Junior counts plies differently than other programs (for example). Simply discount leaf nodes and qnodes and suddenly our node count vanishes to a small number. Perhaps Vas defines a node as 'a fully searched node'.

At any rate, the node count is not actually used for anything except FYI. I really don't understand why people are upset about it.

I can think of just one valid reason: Some are interested in comparing programs, for any of several reasons. And one feature that defines a program's strength is "knowledge". For years everyone has compared their speed to that of Fritz, or genius, or any of a dozen other programs, and used that to draw conclusions. It is no longer possible, since the data is obfuscated to mislead such investigations.

The solution is to decompile these programs, find out what makes them tick.

It's sad to have to do this, but is there any other alternative?

[Dann Corbit]

Multiply by ten (10x). I compared Rybka 1.0 and Strelka 2.0 and average
ratio in node speed was 11,6. So 10 is good estimate in whole game.

Jouni

Why does anyone think that the node count matters at all anyway?

The proof of the pudding is in the eating. Rybka plays strong, that's enough.

Junior counts plies differently than other programs (for example). Simply discount leaf nodes and qnodes and suddenly our node count vanishes to a small number. Perhaps Vas defines a node as 'a fully searched node'.

At any rate, the node count is not actually used for anything except FYI. I really don't understand why people are upset about it.

I can think of just one valid reason: Some are interested in comparing programs, for any of several reasons. And one feature that defines a program's strength is "knowledge". For years everyone has compared their speed to that of Fritz, or genius, or any of a dozen other programs, and used that to draw conclusions. It is no longer possible, since the data is obfuscated to mislead such investigations.

I do not think any motives are proven. Perhaps Vas has simply chosen another notion for the meaning of a node.

The solution is to decompile these programs, find out what makes them tick.

It's sad to have to do this, but is there any other alternative?

Please tell me what this node count is going to do for you. If this is the response that people get, I would suggest that node counts be eliminated.

The response to Rybka's node counts is a bizarre mystery to me. Lots of engines will vary by more than 50% in their node counts. If you looked at a node count of 3 billion verses 30 million, what would those two numbers mean to you?

[bob]

Multiply by ten (10x). I compared Rybka 1.0 and Strelka 2.0 and average
ratio in node speed was 11,6. So 10 is good estimate in whole game.

Jouni

Why does anyone think that the node count matters at all anyway?

The proof of the pudding is in the eating. Rybka plays strong, that's enough.

Junior counts plies differently than other programs (for example). Simply discount leaf nodes and qnodes and suddenly our node count vanishes to a small number. Perhaps Vas defines a node as 'a fully searched node'.

At any rate, the node count is not actually used for anything except FYI. I really don't understand why people are upset about it.

I can think of just one valid reason: Some are interested in comparing programs, for any of several reasons. And one feature that defines a program's strength is "knowledge". For years everyone has compared their speed to that of Fritz, or genius, or any of a dozen other programs, and used that to draw conclusions. It is no longer possible, since the data is obfuscated to mislead such investigations.

I do not think any motives are proven. Perhaps Vas has simply chosen another notion for the meaning of a node.

The solution is to decompile these programs, find out what makes them tick.

It's sad to have to do this, but is there any other alternative?

Please tell me what this node count is going to do for you. If this is the response that people get, I would suggest that node counts be eliminated.

The response to Rybka's node counts is a bizarre mystery to me. Lots of engines will vary by more than 50% in their node counts. If you looked at a node count of 3 billion verses 30 million, what would those two numbers mean to you?

The point is this: node count is proportional to search space. A program with a low (real) node count will behave differently than a program with a high (real) node count, in tactical situations. The same it true for positional tests. And when you look at positional tests and tactical tests, and include node counts, you get an idea of what is being done by evaluation and what is being done by search. That is why Rybka's node count is obfuscated, IMHO.

The computer science literature is full of references to tree searching and nodes, and it _clearly_ defines "node". If we are going to discuss things, and when I talk about a horse I talk about what everyone else calls a horse, but you are talking about what ought to be called a pig, our conversation is not going to carry a lot of information when you try to describe how to clean a "horse" and then cook "hams"...

While what happens within a node might be different between programs, the base definition should not change.

Why one would choose to do that is beyond me... It creates another new language where communication is impossible. Then, what's the point of communicating at all?

[kgburcham]

That is what I was thinking Dan.
Even if Vas adds a zero to the existing eval, or just have the code count something different so that it equals Shredders count for a variety of positions.
Once this is done what will the guys do with this info? Will they go back into the code and see how Vas made the change?
No matter how this plays out, it can still be objected to by some because of the variety of definitions here.

Jibber-Jabber. Vas just do not voice enable Rybka 3. I do not want another talking program. Jibber-Jabber

kburcham

[Dann Corbit]

That is what I was thinking Dan.
Even if Vas adds a zero to the existing eval, or just have the code count something different so that it equals Shredders count for a variety of positions.
Once this is done what will the guys do with this info? Will they go back into the code and see how Vas made the change?
No matter how this plays out, it can still be objected to by some because of the variety of definitions here.

Jibber-Jabber. Vas just do not voice enable Rybka 3. I do not want another talking program. Jibber-Jabber

kburcham

I have an educated guess that Vas and Convetka are looking at fundamentally new paradigms for display of chess information. I guess that every serious chess player will buy one when they are done.

Time will tell.

[M ANSARI]

The node count controversy has not gone unnoticed by Vas. I think that in Rybka 3 a superior way of providing all such information in a useful manner will be available.