Just for fun: IPON BayesELO

Discussion of computer chess matches and engine tournaments.

Moderator: Ras

mcostalba
Posts: 2684
Joined: Sat Jun 14, 2008 9:17 pm

Just for fun: IPON BayesELO

Post by mcostalba »

While I was looking at the world famous IPON list :-) I was wondering, just for pure curiosity, how the list would look if instead comparing 23 engines across more than 400 ELO spawn, the computation of bayes ELO would be done only inside groups of similar strength engines. These group can contain different number of engines but with the constrain that the average ELO of the groups preceding and following a given group are quite separated. For instance a possible distribution could be:

Code: Select all

GROUP 1:

1 Houdini 2.0 STD 
2 Critter 1.4 SSE42  
3 Komodo 4 SSE42           
4 Deep Rybka 4.1 SSE42     
5 Stockfish 2.1.1 JA   


GROUP 2:

6 Chiron 1.1a             
7 Naum 4.2               
8 Fritz 13 32b           
9 Deep Shredder 12          
10 Gull 1.2                 
11 Deep Sjeng c't 2010 32b  
12 Spike


GROUP 3:

13 Protector 1.4.0        
14 Hannibal 1.1           
15 spark-1.0 SSE42        
16 HIARCS 13.2 MP 32b      
17 Deep Junior 12.5       
18 Zappa Mexico II


GROUP 4:

19 Deep Onno 1-2-70         
20 Strelka 2.0 B          
21 Umko 1.2 SSE42 


GROUP 5:

22 Loop 2007     
23 Jonny 4.00 32b   
24 Tornado 4.80     
25 Crafty 23.3 JA
Because the groups are evaluated in a disjoint way it is possible to give an ELO score valid only among one group, so I'd suggest, instead of give the absolute ELO value, to give the difference in ELO points from the first engine of the list, for instance with current values we would have:

Code: Select all

GROUP 1:

1 Houdini 2.0 STD                     0
2 Critter 1.4 SSE42                 -39
3 Komodo 4 SSE42                    -41
4 Deep Rybka 4.1 SSE42              -60
5 Stockfish 2.1.1 JA                -75


GROUP 2:

6 Chiron 1.1a                              0
7 Naum 4.2                                -6
8 Fritz 13 32b                           -14
9 Deep Shredder 12                       -33
10 Gull 1.2                              -38
11 Deep Sjeng c't 2010 32b               -45
12 Spike                                 -48