Quote:
Originally Posted by MJC922
Good morning Class, I haven't noticed drift being an issue with this process. My understanding is drift comes from speed figure makers who are projecting too much improvement or not enough improvement for race winners. What's going on there is kind of a top-down approach.
|
That's exactly where the drift comes from with speed figures.
What I am worried about with my process is that since the ratings are automated, if they have even the slightest bias up or down and those ratings feed future ratings, they will eventually drift up/down noticeably. I've used Class PARs to help anchor them. But the way I do it can cause other occasional inaccuracies I'd like to avoid even if it solves the drift.
I had an experience with drift in the past with automated running style generation. So I want to want to try to avoid that.