Hi,
I'm develloping a website to ELO rate images
The elo system is well-known but the selection of opponents seems to not really be documented
For the second opponent, i choosed a gauss function centered of the elo of the first opponent. Here i can adjust the mu variable
But for the first i have a problem. I tested two methods:
- The first opponent is the element with the least "matchs" (usefull for rating new elements)
- The first opponent is random
So i did simulation and obtain this. I don't knpw what to think, the random one seems t have better results for the same number of matches
Simulation: 1,200 opponents, 10,000 matchs
I'm develloping a website to ELO rate images
The elo system is well-known but the selection of opponents seems to not really be documented
For the second opponent, i choosed a gauss function centered of the elo of the first opponent. Here i can adjust the mu variable
But for the first i have a problem. I tested two methods:
- The first opponent is the element with the least "matchs" (usefull for rating new elements)
- The first opponent is random
So i did simulation and obtain this. I don't knpw what to think, the random one seems t have better results for the same number of matches
Simulation: 1,200 opponents, 10,000 matchs
