LS ratinglist: Robodini 1.1

[pohl4711]

Hi all,

the result of Robodini 1.1 is now online. Because of the stability problems of this engine, it will not become a part of the LS top10-tournament...

http://ls-ratinglist.beepworld.de


Stefan

(Perhaps you have to clear your browsercache or reload the website)

[pohl4711]

If someone is interested in the single results of that testrun:

Robodini 1.1 x64 - Houdini 3 x64 475.5 - 524.5 +231/=489/-280 47.55%
Robodini 1.1 x64 - Critter 1.6a x64 614.5 - 385.5 +378/=473/-149 61.45%
Robodini 1.1 x64 - Strelka 5.5 x64 627.5 - 372.5 +388/=479/-133 62.75%
Robodini 1.1 x64 - Komodo 5 x64 613.0 - 387.0 +419/=388/-193 61.30%
Robodini 1.1 x64 - Robbolito 0.21Q x64s 672.0 - 328.0 +456/=432/-112 67.20%
Robodini 1.1 x64 - Ivanhoe 46eQi x64s 668.0 - 332.0 +444/=448/-108 66.80%
Robodini 1.1 x64 - Bouquet 1.6 x64s 677.5 - 322.5 +470/=415/-115 67.75%
Robodini 1.1 x64 - Vitruvius 1.11C x64s 684.0 - 316.0 +466/=436/-98 68.40%
Robodini 1.1 x64 - Rybka 4.1 x64s 651.5 - 348.5 +463/=377/-160 65.15%
Robodini 1.1 x64 - PanChess 00.250 x64 683.5 - 316.5 +484/=399/-117 68.35%
Robodini 1.1 x64 - Robbolito 0.085g3 x64 709.0 - 291.0 +501/=416/-83 70.90%


Stefan

[Ajedrecista]

Hello Stefan:

Thank you very much for the info! It was indeed very useful for me.

Code: Select all

Robodini 1.1 x64 - Houdini 3 x64           475.5 - 524.5   +231/=489/-280   47.55% 
Robodini 1.1 x64 - Critter 1.6a x64        614.5 - 385.5   +378/=473/-149   61.45% 
Robodini 1.1 x64 - Strelka 5.5 x64         627.5 - 372.5   +388/=479/-133   62.75% 
Robodini 1.1 x64 - Komodo 5 x64            613.0 - 387.0   +419/=388/-193   61.30% 
Robodini 1.1 x64 - Robbolito 0.21Q x64s    672.0 - 328.0   +456/=432/-112   67.20% 
Robodini 1.1 x64 - Ivanhoe 46eQi x64s      668.0 - 332.0   +444/=448/-108   66.80% 
Robodini 1.1 x64 - Bouquet 1.6 x64s        677.5 - 322.5   +470/=415/-115   67.75% 
Robodini 1.1 x64 - Vitruvius 1.11C x64s    684.0 - 316.0   +466/=436/-98    68.40% 
Robodini 1.1 x64 - Rybka 4.1 x64s          651.5 - 348.5   +463/=377/-160   65.15% 
Robodini 1.1 x64 - PanChess 00.250 x64     683.5 - 316.5   +484/=399/-117   68.35% 
Robodini 1.1 x64 - Robbolito 0.085g3 x64   709.0 - 291.0   +501/=416/-83    70.90%

Code: Select all

Draw ratio of Robodini 1.1 against different engines:

Houdini 3:           48.9%.
Strelka 5.5:         47.9%.
Critter 1.6a:        47.3%.
Ivanhoe 46eQi:       44.8%.
Vitruvius 1.11C:     43.6%.
Robbolito 0.21Q:     43.2%.
Robbolito 0.085g3:   41.6%.
Bouquet 1.6:         41.5%.
PanChess 00.250:     39.9%.
Komodo 5:            38.8%.
Rybka 4.1:           37.7%.

If you take a look to this thread you will notice that Kai and me developped two different numerical methods for comparison... and why I say it? Just because two very similar engines A and B are expected to draw much more than two non-related engines A and C (if we assume that B and C are of similar strength, and also A is not far better or far worse than B and C). Adam also developped his own method, but it is too complicated for me; Kai's and mine ones are more easy to implement for me (in theory, my method is the worst of all three). I take the formula of c_i from this post in OpenChess:

Code: Select all

_i stands for subindex i.

µ_i: score of the i-th engine. 
D_i: draw ratio of the i-th engine. 

c_i = (0.5 + |µ_i - 0.5|)*D_i

Kai's method is simply D_i/[µ_i*(1 - µ_i)]. Here are the results I obtained (hoping no typos). So I write opponent, Kai, c:

Code: Select all

Opponent:             Kai       Me

Houdini 3            1.9607   0.1282
Critter 1.6a         1.9967   0.1453
Strelka 5.5          2.0493   0.1503
Komodo 5             1.6355   0.1189
Robbolito 0.21Q      1.9599   0.1452
Ivanhoe 46eQi        2.0201   0.1496
Bouquet 1.6          1.8994   0.1406
Vitruvius 1.11C      2.0172   0.1491
Rybka 4.1            1.6604   0.1228
PanChess 00.250      1.8444   0.1364
Robbolito 0.085g3    2.0163   0.1475

Values are rounded up to 0.0001; for each column, a higher number means that engines draw more in comparison with other engines, taking into account the strength of each engine.

If I normalize those values in this way: x'_i = [sum(x_i)]/x_i (I took this idea from Don):

Code: Select all

Opponent:              Kai'      Me'

Houdini 3            10.7410   11.9611
Critter 1.6a         10.5473   10.5546
Strelka 5.5          10.2769   10.2065
Komodo 5             12.8764   12.8983
Robbolito 0.21Q      10.7452   10.5675
Ivanhoe 46eQi        10.4254   10.2511
Bouquet 1.6          11.0878   10.9111
Vitruvius 1.11C      10.4403   10.2869
Rybka 4.1            12.6833   12.4902
PanChess 00.250      11.4181   11.2490
Robbolito 0.085g3    10.4448   10.4012

Values are rounded up to 0.0001 again; for each column, this time a higher number should mean that engines draw less in comparison with other engines (taking into account the strength of each engine) and one could expect that a higher number in this last code box means that Robodini and i-th engine are more different than others.

I see that there are lots of IPPOLIT derivatives in this match so I expect that the numbers I calculated should be reasonably similar among these derivatives. Komodo is clearly shown as a non-derivative, as well as Rybka. Of course it is not a definitive method for identify similarity of engines, so please do not give too much importance to these results. Sorry for the long post and please keep up your good work!

Regards from Spain.

Ajedrecista.

[velmarin]

Curious the beating that the "monster" gives his "brother Critter".

And produced in two weeks
Bad times for chess

[mar]

Curious the beating that the "monster" gives his "brother Critter".

And produced in two weeks
Bad times for chess

What's bad for chess IMHO are in fact copycats who can only cripple existing "public domain" code and whose best friend is statistical noise;
with absolutely zero understanding of even elementary engine mechanics.

[velmarin]

Curious the beating that the "monster" gives his "brother Critter".

And produced in two weeks
Bad times for chess

What's bad for chess IMHO are in fact copycats who can only cripple existing "public domain" code and whose best friend is statistical noise;
with absolutely zero understanding of even elementary engine mechanics.

Though it clear again.

You miss the first one, because you lack knowledge.


You think they can be always withthe same refrain,
You know perfectly well what is the public domain, so it's okay.
If you persist you are trying to miss.
Teach your stripes, great programmer.

[velmarin]

What's bad for chess IMHO are in fact copycats who can only cripple existing "public domain" code and whose best friend is statistical noise;
with absolutely zero understanding of even elementary engine mechanics.

A message posted to annoy,

but this is supported, if you answer, they cry.

Do not bother and you will be respected.

great Master Martin...

[mar]

Though it clear again.

You miss the first one, because you lack knowledge.


You think they can be always withthe same refrain,
You know perfectly well what is the public domain, so it's okay.
If you persist you are trying to miss.
Teach your stripes, great programmer.

I think I made myself clear. I'm not responsible for your complexes by the way.
It's funny now that you removed your first insulting post
I'm not discussing with guys who delete chunks of conversation, sorry.
Bye Jose.

[michiguel]

Though it clear again.

You miss the first one, because you lack knowledge.


You think they can be always withthe same refrain,
You know perfectly well what is the public domain, so it's okay.
If you persist you are trying to miss.
Teach your stripes, great programmer.

I think I made myself clear. I'm not responsible for your complexes by the way.
It's funny now that you removed your first insulting post
I'm not discussing with guys who delete chunks of conversation, sorry.
Bye Jose.

[MODERATION]
Insulting post (and the branch that followed) was removed by the mod team. There was no enough time to tell members involved.
Miguel

[mar]

[MODERATION]
Insulting post (and the branch that followed) was removed by the mod team. There was no enough time to tell members involved.
Miguel

In that case that I apologize to Jose.