The measure of relative strenght has already been done (see test history): there is about 100 ELO difference between one skill level and the next, indipendently from the hardware and TC, at least until level 10 or something.lucasart wrote:Instead, what would be more interesting is to replace this skill level by the standard UCI_LimitStrength and UCI_Elo. We need to do some measure of the skill levels, and come up with a scaling function that translates them into elo and vice versa.mcostalba wrote:Well, this one is not bad. Indeed "Strength" or perhaps "Strength Level "seems to me a nicer name than "Skill Level"....perhaps we should adopt it...lucasart wrote:He renamed Skill Level into "Strength" and rescaled it by 10.Code: Select all
option name Strength type spin default 200 min 0 max 200
What is more difficult is compare to human ELO simply because human ELO is computed in a much more unreliable and random way: even if you ask someone to play 1000 games vs SF at level 1, he will not be constant during the games and also his ELO level is grossely approximated, say +-100 ELO.