View Single Post
Career Officer
Join Date: Dec 2012
Posts: 695
# 28
08-20-2013, 02:27 PM
Quote:
Originally Posted by masterkeychnk5 View Post
Mhhh i noticed while pugstomping rating goes up significantly fastr then when up against a good team and yourself in the minority.
This is indeed difficult to accurately capture, especially since I still don't have team detection in the client. There is already a dampening factor that should prevent too much point loss if you go up a against a superior team that picks on you way more than on your team mates, but so far there is nothing to punish you for being "too good". It's not entirely trivial to decide if you were simply amazing or the other team was truly bad.

There is data that I can (and intend to) use to reduce the problem a bit. Currently the importance of individual matches for the overall score is given by taking averages with the match duration as weights. I could apply some kind of exponential function to further increase the importance of long matches over short ones; also I could give matches with many kills (close to 30 in total, thus indicating a balanced score) a higher weight in the averaging process; the same goes for matches with a high overall dps/hps score which then indicates that all players were good.

But a different averaging of the individual results doesn't fully solve the problem. While it makes longer/better matches more important, it still leaves the situation where a players decides to only participate in pugstomps - if every match is the same, the weights won't change anything. To do something against that, I could apply penalties for onesided matches, but that would also hurt players who are genuinely good and perform outstandingly even against good opponents and achieve decisive wins.

The last way out which requires much more data than I currently have and which is probably only possible for weekly/monthly lists is to only include matches which look as if they were balanced and simply discard everything else. But that could still punish players/groups who are simply significantly better than everyone else (think: the TSI-QEW 15-0 tournament matches would not have been rated because they were too onesided...).
http://hilbertguide.com