Test tournament: RobboLito 0.08e4

[M ANSARI]

R3 has a linear score against Robbolito with higher time controls or of course with bigger hardware. It is basically R3 with a large part of R3's evaluation still not in. This makes the engine executable very fast and thus it will perform very well against R3 at fast time controls but will lose that advantage at longer time controls. I tried a match at 16 0 on my Octa 4 ghz and ran each on 4 Ghz core and R3 single core came out ahead. At 3 0 and 5 0 Robbolito faster executable gives it the edge. I would think that on much longer time controls R3 will come out well ahead.

I think that Robbolito can be compared to what Rybka had in Winfinder in Rybka 2.2, where Winfinder would be able to find tactical shots sometimes 10X faster than normal Rybka by simply removing a lot of knowledge code and letting the engine just count beans really fast.

[Matthias Gemuh]

R3 has a linear score against Robbolito with higher time controls or of course with bigger hardware. It is basically R3 with a large part of R3's evaluation still not in. This makes the engine executable very fast and thus it will perform very well against R3 at fast time controls but will lose that advantage at longer time controls. I tried a match at 16 0 on my Octa 4 ghz and ran each on 4 Ghz core and R3 single core came out ahead. At 3 0 and 5 0 Robbolito faster executable gives it the edge. I would think that on much longer time controls R3 will come out well ahead.

I think that Robbolito can be compared to what Rybka had in Winfinder in Rybka 2.2, where Winfinder would be able to find tactical shots sometimes 10X faster than normal Rybka by simply removing a lot of knowledge code and letting the engine just count beans really fast.

Actually RobboLito has knowledge that Rybka 3 lacks, e.g. in KBP vs K.
I have not seen any situation yet in which Rybka 3 has more knowledge than RobboLito.

Matthias.

[kingliveson]

R3 has a linear score against Robbolito with higher time controls or of course with bigger hardware. It is basically R3 with a large part of R3's evaluation still not in. This makes the engine executable very fast and thus it will perform very well against R3 at fast time controls but will lose that advantage at longer time controls. I tried a match at 16 0 on my Octa 4 ghz and ran each on 4 Ghz core and R3 single core came out ahead. At 3 0 and 5 0 Robbolito faster executable gives it the edge. I would think that on much longer time controls R3 will come out well ahead.

I think that Robbolito can be compared to what Rybka had in Winfinder in Rybka 2.2, where Winfinder would be able to find tactical shots sometimes 10X faster than normal Rybka by simply removing a lot of knowledge code and letting the engine just count beans really fast.

Not True!

As far as linear score, most top engines behave this way given more time. For example, Zappa Mexico with executable half the size of Rybka's is not an engine that plays blitz well, but you see its strength with longer time control. I have posted the result of 300 15+0 games here where Robbo* came out ahead. I also have 50 30+0 games where Robbo* came out ahead. This may not mean much, but I have 10 90+30 games where RobboLito destroyed Rybka 3. All the games were played with default settings and ponder on.

[garybelton]

From the high depth games I have seen it looks like RobboLito has more knowledge than Rybka in certain positions and less in others, but on balance more knowledge. It's as if the author(s) of it have taken the best bits of many engines (say Rybka, Shredder and Zappa) and rolled them into one. Although the eval of RobboLito is more Rybka-like than say Shredder or Zappa-like so it probably owes most of it's heritage to Rybka. Whoever did it obviously knows about chess engines and may even be a chess engine programmer. Pure supposition on my part of course.

I agree that at lower depths the fact that RobboLito does not have all of the features of Rybka 3 (persistant hash, mpv, code obfuscation, etc) means that it's strength relative to Rybka is greater.

[Uri Blass]

R3 has a linear score against Robbolito with higher time controls or of course with bigger hardware. It is basically R3 with a large part of R3's evaluation still not in. This makes the engine executable very fast and thus it will perform very well against R3 at fast time controls but will lose that advantage at longer time controls. I tried a match at 16 0 on my Octa 4 ghz and ran each on 4 Ghz core and R3 single core came out ahead. At 3 0 and 5 0 Robbolito faster executable gives it the edge. I would think that on much longer time controls R3 will come out well ahead.

I think that Robbolito can be compared to what Rybka had in Winfinder in Rybka 2.2, where Winfinder would be able to find tactical shots sometimes 10X faster than normal Rybka by simply removing a lot of knowledge code and letting the engine just count beans really fast.

I disagree.
I think that robbolito's weakness at long time control is not lack of evaluation but the search.

My guess is that robbolito does too much pruning that is counter productive at long time control.

Uri

[Guenther]

R3 has a linear score against Robbolito with higher time controls or of course with bigger hardware. It is basically R3 with a large part of R3's evaluation still not in. This makes the engine executable very fast and thus it will perform very well against R3 at fast time controls but will lose that advantage at longer time controls. I tried a match at 16 0 on my Octa 4 ghz and ran each on 4 Ghz core and R3 single core came out ahead. At 3 0 and 5 0 Robbolito faster executable gives it the edge. I would think that on much longer time controls R3 will come out well ahead.

I think that Robbolito can be compared to what Rybka had in Winfinder in Rybka 2.2, where Winfinder would be able to find tactical shots sometimes 10X faster than normal Rybka by simply removing a lot of knowledge code and letting the engine just count beans really fast.

Not True!

As far as linear score, most top engines behave this way given more time. For example, Zappa Mexico with executable half the size of Rybka's is not an engine that plays blitz well, but you see its strength with longer time control.
...snip...

Size has nothing to do with being good at Blitz or not...

[Milos]

R3 has a linear score against Robbolito with higher time controls or of course with bigger hardware. It is basically R3 with a large part of R3's evaluation still not in. This makes the engine executable very fast and thus it will perform very well against R3 at fast time controls but will lose that advantage at longer time controls. I tried a match at 16 0 on my Octa 4 ghz and ran each on 4 Ghz core and R3 single core came out ahead. At 3 0 and 5 0 Robbolito faster executable gives it the edge. I would think that on much longer time controls R3 will come out well ahead.

I think that Robbolito can be compared to what Rybka had in Winfinder in Rybka 2.2, where Winfinder would be able to find tactical shots sometimes 10X faster than normal Rybka by simply removing a lot of knowledge code and letting the engine just count beans really fast.

I pretty sure you are just fabricating all these claims and conclusions without even a grain of truth in any of them.
I have bunch of tournaments between Robbo and R3 on half an hour and one hour blitz time control. And when R3 used one core it never scored higher then 42%.
I have also tournaments with 50 and more games where Robbo win against R3 on 4 cores with 10 and 15 minutes blitz control.
And saying Robbo has Rybka's evaluation is exactly the same as saying Rybka has Fruit's evaluation.
Moreover claims about Rybka and Fruit have more weight because they are provable. On the other hand, you cannot prove you claims even when having source code of Robbo.

[ArmyBridge]

R3 has a linear score against Robbolito with higher time controls or of course with bigger hardware. It is basically R3 with a large part of R3's evaluation still not in. This makes the engine executable very fast and thus it will perform very well against R3 at fast time controls but will lose that advantage at longer time controls. I tried a match at 16 0 on my Octa 4 ghz and ran each on 4 Ghz core and R3 single core came out ahead. At 3 0 and 5 0 Robbolito faster executable gives it the edge. I would think that on much longer time controls R3 will come out well ahead.

I think that Robbolito can be compared to what Rybka had in Winfinder in Rybka 2.2, where Winfinder would be able to find tactical shots sometimes 10X faster than normal Rybka by simply removing a lot of knowledge code and letting the engine just count beans really fast.

I pretty sure you are just fabricating all these claims and conclusions without even a grain of truth in any of them.
I have bunch of tournaments between Robbo and R3 on half an hour and one hour blitz time control. And when R3 used one core it never scored higher then 42%.
I have also tournaments with 50 and more games where Robbo win against R3 on 4 cores with 10 and 15 minutes blitz control.
And saying Robbo has Rybka's evaluation is exactly the same as saying Rybka has Fruit's evaluation.
Moreover claims about Rybka and Fruit have more weight because they are provable. On the other hand, you cannot prove you claims even when having source code of Robbo.

Hi Milos could you share the games?
Tnx in advance