Boards›Gameplay

@Stats for MM. I need some clarity on the viewpoint as i don't get it.

Kai Guy·1/13/2020, 5:22:28 AM·2 votes·2,502 views

I don't get it. I don't see any possible way to make them relevant enough outside of the massive noise per game. I absolutely don't understand the players who claim it can be used for MM over the current MMR system that's a derivative of Elo's which is literally a statistical endeavor.

I don't see how individualized champion stats are usable for MM given champion selection happens post lobby. The moment some one randomly picks anything new or swaps to a unpredicted pick It seems like things would get fucked.

I don't see how you can possibly make arbitrary best case usages and judge it objectively that accurately reflect the intricacies of abilities usage. Like using skills for mobility when its situationally helpful. Akali Evelynn Fiora Renekton Nami Kled Riven Rumble can all use skills for mobility gains and incorrect stat analysis makes that usage look substandard even if it was the ideal play. There are other examples such as wave clear.

I just don't see any way to get thru the noise of the numbers with out using Game to game specific subjective judgment. I am aware of no method to automate this with modern tech and yes that includes Machine learning. I cant get a conversation with anyone about the impact between playerskill, Ability's on CD are not a threat so Good players don't mindlessly spam abilities when they are not threatening. Improper skill usage almost always results in higher stats in mobas. Then there's a huge gap between point and click and skill shots and individual skill shots off hitbox size as well as ability range.

The more in-depth and complex you want to make it then the harder it gets. Wanna try and account for champion interaction and synergy? That's a pretty massive # of permutations. I don't know a formula for it, do you?

So, then what about Stat or personal adjustments? Well... its already done? Like Microsoft Trueskill 2 and Blizzard do this. Yet you see the exact same OMG WTF MM BROKEN threads there as in league so what exactly makes it better if you see identical complaints on the topic that can be copy and pasted many times between games with no impact on content? "Fix your shit MM" is a normal QQ thread in any team game.

Look, when I tell people, "Its less effort to pad stats then to try to win games" I am being very literal with you. I am trying to be accurate and honest. I am acting in good faith. Elo's theory works. Its time tested and peer reviewed for about half a century now. The steps needed to convert it to what's commonly called a MMR system are not any mystery.

So... can I get some clarity. Can some one answer in details? I am so very tired of folks waving a magic wand of "Math!" Motherfuckers MATH IS PROVABLE. You list the logic and the steps. You can teach people math. You show formulas, you show the theory, you show examples. You build on existing knowledge. The base assumptions made can be expressed and explained.

13 Comments

Feirisu NyanNyan1/13/2020, 6:51:14 AM4 votes

Hi there, Before I go into details for the summoner 13 that you're asking let us have a quick look at the information we have on matchmaking :

Apparently there are three factors for matchmaking

  • Fair matches - Each team is roughly the same skill
  • Position preference - You get to play a position you want to play
  • Fast queue times - The time spent queuing is as short as possible

and due to stuff like :

  • Matches played during low-activity times of day
  • Parties of players with very different MMRs
  • A shortage of one or more positions (sometimes resulting in one or more players being autofilled)

matchmaker can give high weightage to one of the three factors:

These all throw curveballs at the matchmaker, forcing it to weigh one aspect more heavily than the others.

With MMR hidden the decade old sayings like "MMR is same" doesn't hold true and even with MMR hidden it is blatantly visible that MM ignores Fair matching and position preference for low queue times. I am not going into champ select, champ synergy, player synergy and player behavior (int, troll, afk), but the pre-game lobby where a team is put together.

MMR system is based on Elo system and though Elo is very well established and proven system, the moba games are not chess. **What does it mean? **Individual performances doesn't translate to win/loss and it is a team game. With MMR calculation hidden, all we know is that MMR increases/decreases with win/loss and the change of MMR depends on player's MMR, his/her team's average MMR and opponent team's MMR; ofcourse players keep writing stuff like "play good and you will win more/climb" but ultimately it is a team game and individual performances have limited agency, even for a high impact role like Jungler or Mid. For example, an autofilled jungler is pretty much a loss while an autofilled support isn't having that much chances for a loss, so just from the game lobby "chances" of victory is decided by MM. MMR is the end result and MM is what deciding it most of the times, so treat them as different entities.

MMR system is bad, but not worse than MM system and if anything just by fixing MM system the MMR can reflect the skill of player better. Players raging "fix MM" or "omg fk RITO broken MM" are all pointing at the bad MM system only and this is why even with META champs, many games are lost because MM decided it at start itself. Having high winrate(50+%)? or in promos translates to getting more autofills on team and inters, trolls, afk on team because MM wants to make sure that everyone is forced to play more games to climb on average. Don't write "how RITO can detect this", because they actually have PRA(Performance based Rank Adjustment), but use it in negative direction. Instead of matching "good performing" players with "good performing" players, they do it in "good performing"+"bad performing" players and it is easy to detect "good performing" from "bad performing" players from past data. This indeed fails for new accounts and placing them in low gold games is entirely different issue.

MATH IS PROVABLE. You list the logic and the steps. You can teach people math. You show formulas, you show the theory, you show examples. You build on existing knowledge. The base assumptions made can be expressed and explained.

Except that there is no math or formulae or logic for the current MM system [slayer-jinx-wink]