a response to anthony Cozzie opinion on rybka

[Alessandro Scotti]

as Cozzie stated, there isn't even anything "new" in rybka/strelka - so Hyatts has no point at all.

It's fun and just a little bit hopeless that this kind of statement happens every single time the "secrets" of a strong engine are revealed.
When Fruit came out, possibly the most innovative engine in the last ten years or so, lot of people hurried to claim exactly that: there was nothing new in Fruit.
Today, my guess is that many (most?) of the strongest engines out there are based on the nothing new that was in Fruit...

[Uri Blass]

The most interesting thing in Strelka is the material imbalance lookup tables. These come from this:
http://home.comcast.net/~danheisman/Art ... alance.htm
http://home.comcast.net/~danheisman/Art ... _pawns.htm

It is perhaps not surprising that Larry Kaufman is on the Rybka team.

Although Mr. Kaufman published his results a long time ago, I do not think anyone picked up the idea to put this into a chess program until the Rybka author bothered to do it. It is clearly a very, very good idea.

I do not think that this idea is the main reason for the fact that strelka is strong.

Without material imbalance tables strelka is weaker but the main reason is simply the fact that it has wrong evaluation of pieces and overevaluate bishop and knight relative to rook and pawn.

I guess that if you drop the material imbalance tables and correct the value of pieces then strelka is not going to lose more than 30 elo rating points.

Uri

[Uri Blass]

as Cozzie stated, there isn't even anything "new" in rybka/strelka - so Hyatts has no point at all.

It's fun and just a little bit hopeless that this kind of statement happens every single time the "secrets" of a strong engine are revealed.
When Fruit came out, possibly the most innovative engine in the last ten years or so, lot of people hurried to claim exactly that: there was nothing new in Fruit.
Today, my guess is that many (most?) of the strongest engines out there are based on the nothing new that was in Fruit...

It is funny.
I totally disagree that there is nothing new even if every detail in the code is based on old ideas.

Imagine the following situation.

Program AB use ideas A and B.
Program AC use ideas A and C

programmer X writes
Program BC that use ideas B and C and BC is significantly stronger than AB or AC.

Is it correct to say that X discovered nothing new?

I think that it is not correct.
X discovered that ideas B and C work together and it was not something obvious because other people did not find it.

Uri

[geots]

Graham, the mods stand beyond criticism with decisions, but they should listen to what some have to say about the underlying topic IMO. I for one didnt see the now removed messages but I want to make a remark on this thread.

If you read carefully the message from "Joseph" you will find a quote from Anthony. And even I as a foreigner can understand that this sort of incomprehensible (for most) humor is in itself a critic. I do hope that with your deletions you didnt delete fairly legal discussion of the statement from Anthony, as it's quoted by "Joseph".

It's self-understood, that not everybody must agree with the opinions in the statement quoted from Cozzie, or do the mods disagree here? So, why deleting the opinions?

Let me add an unwanted IMO opinion. Since nobody really knows what Vas did in RYBKA, some spys are always active and are trying to find out something. And I tell you something, I admire the patience of Vas and that he keeps out of such hanky panky.

With all due respect Rolf, if you didn't see the removed posts, there's not much point commenting on the decision.
I repeat that CCC is not the place for character dissections.

Regards, Graham.

Please - I didnt tear this into doubts. What I asked you was if you mods didnt realise that what was quoted from Zappa author, is also a bit in direction of character dissection, no? It's a bit a sort of indirectly verbalised putting down. But you are better in this language.

Rolf, i understand the English language very well- you can count on it. And you hit the proverbial nail directly on its proverbial head. BTW, i know you have seen many decks of cards. Do you know why you only see one of the jacks' eyes? (Read quickly- my thread wont be here any longer than i will)

[Guetti]

The most interesting thing in Strelka is the material imbalance lookup tables. These come from this:
http://home.comcast.net/~danheisman/Art ... alance.htm
http://home.comcast.net/~danheisman/Art ... _pawns.htm

It is perhaps not surprising that Larry Kaufman is on the Rybka team.

Although Mr. Kaufman published his results a long time ago, I do not think anyone picked up the idea to put this into a chess program until the Rybka author bothered to do it. It is clearly a very, very good idea.

I do not think that this idea is the main reason for the fact that strelka is strong.

Without material imbalance tables strelka is weaker but the main reason is simply the fact that it has wrong evaluation of pieces and overevaluate bishop and knight relative to rook and pawn.

I guess that if you drop the material imbalance tables and correct the value of pieces then strelka is not going to lose more than 30 elo rating points.

Uri

During CCT there was a discussion about Strelkas secrets between Anthony and Vasik going on. If I understood Vasik correctly, then the tables were completely removed/replaced in newer versions of Rybka.

[Dann Corbit]

The most interesting thing in Strelka is the material imbalance lookup tables. These come from this:
http://home.comcast.net/~danheisman/Art ... alance.htm
http://home.comcast.net/~danheisman/Art ... _pawns.htm

It is perhaps not surprising that Larry Kaufman is on the Rybka team.

Although Mr. Kaufman published his results a long time ago, I do not think anyone picked up the idea to put this into a chess program until the Rybka author bothered to do it. It is clearly a very, very good idea.

I do not think that this idea is the main reason for the fact that strelka is strong.

Without material imbalance tables strelka is weaker but the main reason is simply the fact that it has wrong evaluation of pieces and overevaluate bishop and knight relative to rook and pawn.

I guess that if you drop the material imbalance tables and correct the value of pieces then strelka is not going to lose more than 30 elo rating points.

Uri

I do not know why Strelka is strong. I do know that the material imbalance idea is unique to Strelka and Rybka (so far). I guess that there are other unique ideas as well, but I did not diagnose them yet.

[frosch]

If there is nothing new in Strelka, tell me who the others are who implemented Larry Kaufman's ideas? Answer:zero. Oops. One: Rybka

kaufman only does work on the evaluation funktion. moreover he started with that only before the rybka 2.3.2a release. his main work will be in rybka 3. it's all very chess related - not that interesting from a scientific point - no complex algos..
but anyway, don't we talk about rybka 1.0beta? IF rajlich used many ideas from others and developed them, there should be something in his first beta.
and strelka is totally based on rybka 1.0beta - I won't argue about that, because it's just so clear. he shortly after that beta announced to go commercial and developed all on his own. I am still puzzled by statements like: more power to the code thiefs..

[Dann Corbit]

If there is nothing new in Strelka, tell me who the others are who implemented Larry Kaufman's ideas? Answer:zero. Oops. One: Rybka

kaufman only does work on the evaluation funktion. moreover he started with that only before the rybka 2.3.2a release. his main work will be in rybka 3. it's all very chess related - not that interesting from a scientific point - no complex algos..
but anyway, don't we talk about rybka 1.0beta? IF rajlich used many ideas from others and developed them, there should be something in his first beta.
and strelka is totally based on rybka 1.0beta - I won't argue about that, because it's just so clear. he shortly after that beta announced to go commercial and developed all on his own. I am still puzzled by statements like: more power to the code thiefs..

Evaluation is just as interesting as search for a chess program. Larry Kaufman shows (for instance) that the a and h pawns are less valuable than any of the others. If your chess program does not take this into account, then it will not play as well as one which does take that idea into account.

His idea was actually very interesting scientifically. He took a large database of chess games and filtered so that the players were close in strength, and then examined the games for material imbalance. Based on the data that he saw, he created a score for any type of material imbalance.

I guess that there are lots of things in his lookup table that we would like to understand.

Though Kaufman started later, it seems clear that Mr. Rajilch used his ideas before he joined the project.

Algorithm complexity is unlikely to be closely coupled to program strength. The strongest programs are simple ones.

Look at the tiny size of Strelka and Thinker. And yet they are very strong. So size or complexity do not have any direct relationship to strength.

I guess that it is like Einstein and Newton. Look at their ideas the way that they explained them. Really very simple. And yet the clarity of genius is really taking the simple idea and understanding it correctly. Before Einstein talked about a man walking on a train or standing in an elevator, we did not understand his ideas. After we heard it, "WOW! How simple!" but if it was really so simple why didn't it ever occur to us?

[mjlef]

The most interesting thing in Strelka is the material imbalance lookup tables. These come from this:
http://home.comcast.net/~danheisman/Art ... alance.htm
http://home.comcast.net/~danheisman/Art ... _pawns.htm

It is perhaps not surprising that Larry Kaufman is on the Rybka team.

Although Mr. Kaufman published his results a long time ago, I do not think anyone picked up the idea to put this into a chess program until the Rybka author bothered to do it. It is clearly a very, very good idea.

I do not think that this idea is the main reason for the fact that strelka is strong.

Without material imbalance tables strelka is weaker but the main reason is simply the fact that it has wrong evaluation of pieces and overevaluate bishop and knight relative to rook and pawn.

I guess that if you drop the material imbalance tables and correct the value of pieces then strelka is not going to lose more than 30 elo rating points.

Uri

I do not know why Strelka is strong. I do know that the material imbalance idea is unique to Strelka and Rybka (so far). I guess that there are other unique ideas as well, but I did not diagnose them yet.

Material Imbalance is not a new idea. You can see commented out information in versions of Toga, and it exists in older versions of Glaurung, for example, using values from Kaufman's papers. Rybka/Strelka has refined that a lot, with a lot more terms and I assume data from games. How Vasik came up with the values is the mystery, but I certainly have played with the idea of analyzing game outcomes for score corrections based on win percentages. So maybe that is the basic idea.

I think it is worth about 100 ELO overall, but note the Strelka material tables have a lot of information in them already present in other programs (like KBK is a draw). With corrected piece values and Fruit-like rules for this stuff, my guess it is might be worth about 50 ELO. Perhaps the newer versions of Rybka have increased that though. In any case, a fast table lookup is just going to be a lot faster than a bunch of rules you apply every time the material changes. When I took a bunch of statistics and came up with a material adjustment in my program, it was worth about 40 ELO, using a rather crude first attempt. Of course, I already had lots of rules for scaling piece values based on material.

I agree with Anthony in general. Strelka (and I assume Rybka) is fast. It packs a lot of knowledge into a fast program. Anthony does not mention the differences between Strelka and Fruit on things like mobility. Strelka uses different values...how much this helps I have not tested. Some see counterintuitive, like reversals in values from opening to endgame compared to Fruit. These might matter more than we expect. It is clean, fast and small code. I assume this small size of the actual code helped in reverse engineering Rybka.

Mark

[Uri Blass]

The most interesting thing in Strelka is the material imbalance lookup tables. These come from this:
http://home.comcast.net/~danheisman/Art ... alance.htm
http://home.comcast.net/~danheisman/Art ... _pawns.htm

It is perhaps not surprising that Larry Kaufman is on the Rybka team.

Although Mr. Kaufman published his results a long time ago, I do not think anyone picked up the idea to put this into a chess program until the Rybka author bothered to do it. It is clearly a very, very good idea.

I do not think that this idea is the main reason for the fact that strelka is strong.

Without material imbalance tables strelka is weaker but the main reason is simply the fact that it has wrong evaluation of pieces and overevaluate bishop and knight relative to rook and pawn.

I guess that if you drop the material imbalance tables and correct the value of pieces then strelka is not going to lose more than 30 elo rating points.

Uri

I do not know why Strelka is strong. I do know that the material imbalance idea is unique to Strelka and Rybka (so far). I guess that there are other unique ideas as well, but I did not diagnose them yet.

I do not know what is new in strelka because I do not know other programs well.

I only know that there are ideas that are new to me.

For example detection of unstoppable passed pawns in the endgame also include cases when the pawn is supported by the king and not only cases when the opponent king is not in the square of the pawn(not that it is the secret for strelka's high playing strength).

Strelka plays c5 is a winning score for white at depth 1.
fruit or glaurung cannot see that the pawn is unstoppable

New game - Strelka 2.0 B
[d]3k4/1K3ppp/8/8/2P5/8/6P1/8 w - - 0 1

Analysis by Strelka 2.0 B:

1.c4-c5
+- (6.38) Depth: 1 00:00:00
1.c4-c5 Kd8-d7 2.c5-c6+ Kd7-d6
+- (6.18) Depth: 2 00:00:00
1.c4-c5 Kd8-d7 2.c5-c6+ Kd7-d6
+- (6.18) Depth: 3 00:00:00
1.c4-c5 Kd8-e7 2.Kb7-c7 Ke7-e6
+- (6.19) Depth: 4 00:00:00
1.c4-c5 Kd8-e7 2.c5-c6 Ke7-e6 3.c6-c7 f7-f5 4.c7-c8Q+ Ke6-e5
+- (8.09) Depth: 5 00:00:00

Uri