Regarding the game-based auto-balancer.

Discussion in 'General' started by Sgt.Security, May 26, 2019.

  1. Sgt.Security

    Sgt.Security Member

    Likes Received:
    Trophy Points:
    I know people've been complaining about our game-based auto-balancer for months, so I've been paying attention specifically to it for the past couple of months.

    And I noticed quite a few interesting things.

    Conclusively, the system actually isn't that bad, there's actually some outside factors that made the system look worse than it actually is.
    So, what are the outside factors?

    (1) False assumptions of how the system works.
    There's been a myth saying that the auto-balancer is using a very crappy method of "1 vet = 2 nub" to balance. This is wrong, I think it's an analogy being taken as fact.

    The actual system is more complicated than that.

    (2) People can freely join spec/quit at any given time.
    So, say, if we have a 9v10, and someone on the "9" side decided to join spec because he's probably on a team that he doesn't like. Then we get a 8v10 (and if people happen to check scoreboard at this time, people'd see 8v10 and complain immediately)

    I swear, usually when this happens, our auto-balancer actually says that the team with less players is the weaker side, that is, the system KNOWS it's stacked, the system is not turning a blind eye and say "this 8v10 is balanced", it's merely that nobody is joining.

    (3) People give up upon seeing unbalanced numbers.
    Say, we have a 19v22, and the 19 side actually has better players.

    The 19 side sometimes, just complain and refuse to play "at their skills".

    These games do end really quickly and people'd say they are horribly balanced, ignoring the fact that they gave up a fair fight.(at least fairer than they thought.)

    Or, sometimes, worse, they just quit and make the numbers look even worse. (18v22..etc)

    (4) People assume !predict gives the perfect answer.
    That is, whenever !predict gives non-50/50 result, people immediately jump to the conclusion that teams are definitely not balanced. Ignoring the fact that !predict is solely based on another data base that has inevitable errors too.

    !predict is merely a reference. It's not, and won't be the perfect answer.

    (5) People ignore when auto-balancer gives their "ideal" games.
    I'll be straight, this happens quite often.

    Nobody ever says the auto-balancer is doing the job when it gives balanced teams in terms of both number and skill.

    These are the "ideal" games for you guys, but they are usually taken for granted and never for once I've heard, even ironically, that people say the system is doing good.

    (6) Vets starting fresh again.
    Clearly, these players would be unregistered and the system would recognize them as new players.

    When you get just 2~3 players like these, it just inevitably screws up the balance a bit.

    This happens in all games and all ranking systems. This is not something we can prevent.

    Now, I am not just going to blame it all on outside factors, here's the problems that the system actually do possess.

    (1) It does ignore player count difference.
    The system currently only looks at skill.

    And I agree that, if it's a 19v22, then the 19 side does possess a bit of disadvantage EVEN IF the teams are actually skill-wise balanced. We can't say the 3 missing players don't make any difference, that's 3 less revive engineers or maybe just 3 sandbags, in late game that could be 3 less tanks.

    Our system currently ignores this, and this will be considered soon.

    (2) The database still needs a bit of time to brew itself.

    I have access to the database and I can assure you that it has improved greatly. I know each and everyone of you down to details and I can view your skills from the top, I know how good you are, and I think our database is actually slowly climbing to my expectations.

    Just give it a bit more time.

Share This Page