Example:
Komodo have 100 games vs. Stockfish vs. all others the 50 games.
Gaviota have 100 or 150 games vs. a small number of opponents.
Good idea!
In my opinion:
The higher the number of games and the higher the number of opponents and the lesser the error of opponents the more exactly will be the own / final result in Elo.
That is the point!
I will view that in my next versions of my Rating list and will wrote a message for my news about it.
Means:
For 1:0 for 0:1 and for 0.5:0.5 the move average.
Made sense for one eng-eng match or if People have a database of eng-eng games without resign mode. So we can see in additional nice information we can used for further statistics.
michiguel wrote:I added two of your suggestions
12: OppErr Average of the opponent errors
13: OppN Number of opponents
Which inspire me to add one that I think it is better than OppN
14: OppDiv
Diversity of opponents. If the number of games is equally distributed, it is the same as OppN, but if most of the games come from one opponent, it will get closer to 1. This is the "Effective number of opponents".
Example:
Komodo have 100 games vs. Stockfish vs. all others the 50 games.
What is the exact formula that you use for OppDiv? I find this option quite useful but I could not reproduce the value obtained by Frank. I tried the formulae from the following link:
michiguel wrote:I added two of your suggestions
12: OppErr Average of the opponent errors
13: OppN Number of opponents
Which inspire me to add one that I think it is better than OppN
14: OppDiv
Diversity of opponents. If the number of games is equally distributed, it is the same as OppN, but if most of the games come from one opponent, it will get closer to 1. This is the "Effective number of opponents".
Example:
Komodo have 100 games vs. Stockfish vs. all others the 50 games.
What is the exact formula that you use for OppDiv? I find this option quite useful but I could not reproduce the value obtained by Frank. I tried the formulae from the following link:
Which inspire me to add one that I think it is better than OppN
14: OppDiv
Diversity of opponents. If the number of games is equally distributed, it is the same as OppN, but if most of the games come from one opponent, it will get closer to 1. This is the "Effective number of opponents".
That seems more in the spirit of the request.
Now for mine, regarding players with few games. It's a bit inconvenient, that you have to use the -g switch when some players are under-represented. It should be possible to specify the minimum number of games for a player to be rated, and after the program runs, Ordo could list those that were left out and how many games they had. In a related request, I would also include an option for the user to specify the percentage of games a cluster needs, to be rated. Below that mark, I wouldn't rate games from marginal clusters.