Long contest rating

It seems to me that the rating of long contests doesn’t make sense. For example, in the just now finished JAN13, ACRush beat mugurelionut by the tiny margin of 0.006 (or 1 place, if that’s what counts). His already very high rating (~18400), however, has increased by a whopping 2500 points, while mugurelionut’s, who was rated much lower (~4500) has only increased by 1500.

Shouldn’t contest rating work in reverse, similarly to how ELO works? If you already have a high rating, winning should gain you a smaller increase (and, consequently, losing - a larger decrease) than someone rated lower.

9 Likes

I think that is not exactly how it works. So far I couldn’t find a page explaining how long contests rating works but I can assume certain things. I think contest rating works in reverse when there’s a certain user user_A with rank_A that beats a user_B with rank_B considering that before the contest the situation was rank_B>rank_A, of course this may remain true after the contest but the user with lower ranking will definitely get more points comparing to how many points the user with the higher ranking would get if he beat the later. And I don’t know if long contests ratings work like short ones do but if long contest ratings also have a bonus factor that is calculated the same way that short ones do it would explain why ACRush got so many points. Remember that he finished FIRST and first place is a big deal and if in fact there is a bonus factor like the one from short contests ACRush’s bonus would be a lot higher than mugurelionut’s considering that:

bonus_factor = n/ranks
where n is the number of participants and rank is the rank the user in that contest
for example for a contest with 3000 participants
the user placed first in the contest would get a bonus factor of 3000/1 = 3000
the user placed second would get a bonus factor of 3000/2 = 1500

and that is a big difference; also don’t forget that mugurelionut didn’t beat ACRush so the points he got were calculated using the global rankings of the users that were ranked below him in the contest and considering that not everyone participated there may or may not be so much users that were below his global ranking. Sorry if I couldn’t make myself clear on that part.

But those are all assumptions and I would actually like to see a page explaining how long contests ratings work. And I’m still trying to improve my English so don’t judge too much :);

Yes, even I think it requires modification. I feel it favors people who are participating regularly(including me). Or may be we need to wait for some more contests for it to stabilize.

  • The rating increase should not be linearly (inversely) proportional to the rank. 1st place is not twice better than 2nd place. Maybe sqrt or ln.

  • The score difference should be taken into account, for obvious reasons. Had ACRush beaten 2nd place by 0.5 points or more, it could’ve been considered a significant advantage. A 0.006 point difference could just mean he ran his randomized algorithm a few more times to get a better result.

  • In a long contest, finishing first should not matter.

1 Like

http://blog.codechef.com/2012/03/11/3rd-anniversary-ends-with-a-3-way-tie/
This states that the same rating system will be used for the Long contest which is being used for the short contest.

Details of the rating system of short contest
http://blog.codechef.com/2010/12/27/a-surprise-in-store-this-new-year/

1 Like

@msasha the challenges are supposed to be tie breakers. Read carefully the link that @mukulgupta gave. The calculations are mostly based on the rank you got on the contest. The points are just a way to define your rank during the contest.