1. by RB teh Warboss Choppa on September 10, 2012

    Reported

    The ELO ranking system should not match veteran bad players with unranked new players. New Players which are potentially at an “Expert” skill level create a frustrating gaming environment for veteran players of a very low skill level.

    YOU MUST LOG IN TO ENABLE VOTING & COMMENTING
    ONLY ONE VOTE PER INDIVIDUAL STEAM ACCOUNT FOR THIS ISSUE. ALL VOTES ARE FINAL.

    The ELO ranking system should not match veteran bad players with unranked new players. New Players which are potentially at an “Expert” skill level create a frustrating gaming environment for veteran players of a very low skill level.

    GET MORE VOTES: DISCUSS THIS ON FACEBOOK, REDDIT, TWITTER, AND GOOGLE+
    PRIORITIZE ISSUE FOR VALVE: THUMBS UP (fix), NO VOTE (not important right now), OR THUMBS DOWN (do not fix)
    VN:R_U [1.9.22_1171]
    Rating: +19 (from 27 votes)

    Steps to Reproduce the Problem:
    1. “Veteran” Bad Players (with a really low skill level) are routinely matched against New Players.
    2. A New Player with exceptional skills is matched against this “Veteran” Bad Player (with a really low skill level).
    3. Repeat this scenario for every day a highly skilled New Player uses Valve’s matchmaking.

    Expected or Desired Behavior:
    All new players should have an ELO score between 26% to 46% of the maximum possible ELO score after playing 10 rounds.
    All “Veteran” Bad Players of a really low skill level will essentially out-rank all New Players, and escape the vicious cycle of being matched against expert New Players (on a daily basis).

    VN:R_U [1.9.22_1171]
    Rating: +19 (from 27 votes)

  2. Lash
    about 155 days ago

    VN:R_U [1.9.22_1171]
    Rating: +1 (from 1 vote)

    I’m sure something like this is already being done. The initial skill estimate is that the new gamer will be of mid range ELO/MMR, since the MMR system doesn’t have any other information. The matchmaking will try to match players with similar MMR skill numbers AND uncertainty associated with that estimate. Valve probably modify it a little so that if they do get placed with players with less uncertain “veteran” players, it will be slightly lower than the middle MMR since new players to matchmaking will tend to be lower skilled than the average MMR in find game -> competitive classic.

  3. Azzu
    about 154 days ago

    VN:R_U [1.9.22_1171]
    Rating: 0 (from 0 votes)

    I see what you’re trying to say, Lash, but this issue is about having a better initial skill estimate. Noone doubts that, as soon as you started playing, you will be ranked more appropriately. The issue in this ticket is only about the first few games.

    I find it rather impossible to implement, so in my opinion it doesn’t really make sense to have this ticket, but the wish for it is still valid and the impossibility to have a better initial skill estimate has not been proven.

  4. DeepII
    about 154 days ago

    VN:R_U [1.9.22_1171]
    Rating: 0 (from 0 votes)

    A way to do that would be to use the time someone played either 1.6 or Source. If someone has >1000hrs of either game he’ll probably dominate during his first matchmaking sessions. So this could be an aspect to match someone intially in a more fitting skill group.
    Someone coming from a different FPS could of course still dominate, but he won’t be very familiar with how CS works, so matching him with aim-wise worse players would still be fine.

    1. Tags balance added; elo ranking removed.

  5. Lash
    about 153 days ago

    VN:R_U [1.9.22_1171]
    Rating: +1 (from 1 vote)

    The problem boils down to, how do you model the prior skill of the player as some probability distribution.

    How it is implemented comes down to modelling effort vs. user experience benefit.

    For the first few games, the system at the moment, just like LoL with it’s 5v5 implementation for ranked, will try to first match you up with similar MMR, but more importantly, with similar UNCERTAINTY as well. Often the games you get will be with other players with a similar amount of uncertainty in their skill estimate. This means “veteran” players should not be part of the equation unless there isn’t anyone else there, which then comes down to balancing queuing times with the fairness of the match.

    Ideally, queue times should be less of an issue, but since queue times impact heavily on the user experience, it must be reduced at the cost of finding the fairest match. As I tend to conclude with this problem, if queuing was less of an issue, the user experience in all respects would be improved. So this problem could indirectly be solved to a degree with this (#790)

    @DeepII: Your suggestion of hours played of previous CS has some merit, but coming up with that model and uncertainty in that model needs data. It’s probably correlated though. But will it be any better than the present model they use without this information? Perhaps not by much. Playing more probably improves aim, but it could all have been in 16v16 servers and gun game servers, which means they are no better than the other FPS players. So accommodating hours played may not be that informative. You can’t use K/D in pub servers, as that is biased to the unknown skill level of teammates and opponents.

  6. Azzu
    about 151 days ago

    VN:R_U [1.9.22_1171]
    Rating: 0 (from 0 votes)

    [This does not affect the balance, and "elo ranking" is a valid subcategory of "matchmaking".]

    1. Tags elo ranking added; balance removed.