Skip to content

Mmr calculation

Firestorm-253 edited this page Jan 5, 2023 · 17 revisions

Mmr calculation rules

  • Player ratings serve as an indicator and do not represent a very accurate strength of a player.

  • Our Mmr system is not a "grind" system. The rating represents the current skill of a player independent of the number of total games. After like ~100 games the rating should be reliable.

  • Player ratings can change drastically if another player uploads all of their replays. This player rating will be adjusted, as well as all ratings of players with/against this player.

  • Your apps local Mmr is most likely different from the websites Mmr, which includes all replays in the calculation.

  • If you lose against a very strong team, you lose very few rating points, but if you win, you gain a lot. Because of this, there's no reason to leave a lobby, and shuffling players could backfire. "You never lose. You either win or you learn."

  • The Mmr calculation also depends on the matchup. So if you lose in a very bad matchup (e.g. Horner vs Tychus) you lose fewer points but get more if you win.

  • Leavers (a player who leaves the game without pressing the Victory/Defeat button) receive a penalty in their rating. (Maybe check if the team left at some point to process inhouse defeats)

  • Players who do not upload a lost replay, but that replay is uploaded by one or more other players in the game will receive a penalty in their rating.

  • The MMR will be seasonal. (currently we think about 6 Months, up for testing) Alternatively there will be some kind of decay over time to prevent Players from camping on a Rating.

  • In any case, don't feel pressured while using the app. If you are too afraid of losing rating points, you should create a second account and always play on the account with the lowest rating (switch to the other once you beat the rating).

Features

  • The backbone of the MMR-system is a very popular formula called "ELO". It compares 2 teams strength and based on that your mmr-change is calulated.

    • win vs worse enemies = small MMR gain
    • win vs better enemies = big MMR gain
    • loss vs worse enemies = big MMR loss
    • loss vs better enemies = small MMR loss
  • Consistency:

    • Every player gets a value from 0 to 1. (start at 0)
    • The Consistency in-directly influences how much a player's rating changes:
      • big Consistency = small MMR change
      • small Consistency = big MMR change
    • The Consistency increases and decreases after each game accordingly:
      • if the match result equals the expectation of who will win = increase
      • if the match result doesn't equal the expectation of who will win = decrease
    • It is meant to smooth the rating and decrease big jumps for unexpected wins or losses.
  • Confidence:

    • Every player gets a value from 0 to 1. (start at 0)
    • The Confidence directly influences how much a player's rating changes:
      • big Confidence = small MMR change
      • small Confidence = big MMR change
    • The Confidence increases and decreases after each game accordingly:
      • if the difference of the expected Result and the actual Result (1=win, 0=loss) is bigger than the current Confidence = increase
      • if the difference of the expected Result and the actual Result (1=win, 0=loss) is smaller than the current Confidence = decrease

Features to come

  • (matchup dependent) => The MMR calculation will depend on the matchup. So if you lose in a very bad matchup (e.g. Horner vs Tychus) you lose fewer points but get more if you win.
Clone this wiki locally