Page 1 of 1

Calibrating AI Level for RL based drivers

Posted: Wed Oct 22, 2014 4:03 pm
by ivaneurope
I'd like to know how do you adjust the GR AI Level for Real World based drivers? I tried several methods, but they didn't work.

Re: Calibrating AI Level for RL based drivers

Posted: Wed Oct 22, 2014 4:06 pm
by RobertRacer
Maybe manually?

Re: Calibrating AI Level for RL based drivers

Posted: Wed Oct 22, 2014 6:25 pm
by puttz
I usually take the points leader/season champion and set him as the base of 100. Then set the other drivers based off of that depending on how close they are in the standings, they may stay at the same level as the previous driver or decrease by 1 or 2 (occasionally more). I don't have a definitive formula, I just kind of make judgments on how much to decrease the next driver as I go. Take a look at the AI's included in my NASCAR Sprint Cup 2013 pack to get an idea of how I do it.

I don't know that there really is a good formulaic method that can be used to set AI for real world drivers. Anyone else have a method they use?

Re: Calibrating AI Level for RL based drivers

Posted: Wed Oct 22, 2014 7:46 pm
by kuba
A long time ago I used to drive with a friend in dorm - there were 2 of us, 4 AIs and we had tracks of F1 season from that year and some formula 1 car, I don't remember which one. We had point system 9-6-4-2-1 and 1 point for FL.

We had started with AI with skills 100-97-93-90 and if one of us won the season we increased all IA skills by 1, else we decreased by 1.

Re: Calibrating AI Level for RL based drivers

Posted: Wed Oct 22, 2014 7:52 pm
by ivaneurope
What I tried to use as a calculation method was to have AI 60 as base and then increase the AI under 4 criterias - Podium finishes, Point Finishes, Race Finishes and Pole Positions+Fastest Laps using the formulA (Criteria covered/No. of races)*10. But found this formula as inconsistent for backmarkers - for instance if I use the stats from last season's F1 Max Chilton, who finished all races, will have AI 70, while drivers with points like Bottas and Gutierez(I am talking about 2013) will have AI 68-69. And drivers like Hamilton and Raikkonen would have better AI than Alonso and Webber.

Re: Calibrating AI Level for RL based drivers

Posted: Thu Oct 23, 2014 12:06 pm
by Gzehoo
I made on my computer some F1 seasons in GR with 22 CPU players (every with skill 100 at the start; there were 4 groups with 5-6 drivers on grid) and for every next season I increased them skill level depending on their number of points collected during the season. For example:

>300 points - +6 for skill level
>200 - +5
>150 - +4
>100 - +3
>50 - +2
>0 - +1

I made also a simply rule: basing on last 3 seasons, I took best 2 seasons and changed for each AI his skill. Sometimes it caused that AI with skill 112 was better than AI with skill 104 although this 'better' one had a little worse car.