General DiscussionWhat does Dota 2 Hidden rating depend on?

You're going to have problems with chicken-or-egg... So you've noticed win rate has to do with it. But is it just the case that most high mmr have high win rates, or was the win rate used to compute the high MMR? Same thing with KDR.

An admin on the dota developer forum once said that if you value your MMR, don't play with players you know are far worse than you. It's a safe bet that your MMR is largely influenced by final results (W/L) (as implied by that statement). But that doesn't translate purely into high win ratios since the quality of the win will surely be calculated somehow.

Going on pure conjecture, I'd assume GPM, XPM, last hits, denies, wards purchased, and couriers purchased could be used to better a person's skill approximation. If they don't use these metrics for well-sampled individuals (e.g. 800 wins), they surely do to expedite the detection of smurf accounts in the first handful of games.

Something they do not do, which I would do, is maintain an MMR for each hero played, computing an overall MMR based on the most frequent heroes played. In other words, if I am a grade-A support, having super high skill on all supports and those games accounting for 95% of my games, the 5% of times I am crushed while playing Spectre shouldn't influence my overall MMR as it is a statistical anomaly. The algorithm could be quite advanced, but it could be as simple as doing a weighted average of all MMRs, weighted by the frequency of times that hero is played. E.g. suppose there are only 3 heroes, and I have the following MMRs (on a scale from 0.0 to 1.0) for them:
0.94 0.20 0.002

But out of my 1000 games, I've played:
950 30 20

then my MMR would be
[(950)(0.94) + (30)(0.20) + (20)(0.002)]/(1000) = 0.89904

So despite blowing with the majority of players, I deserve to be in a pretty high rank since I normally don't impact the game with those heroes.

Thank you for your opinion ^^ Your theory is pretty cool.. Anyone else ?

It's determined based on your performance each game. Depends on your performance on the hero compared to the average stats of the hero. If you perform better than the average of the hero, then you get to a higher skill level.

hidden rating doesnt matter because match maker is literally broken Go check swiftendings thread from today or dig up lie to me agent fag that has few thousands replies with some m ids

I originally (a year ago) thought it was going to be purely a TSR type system....so it would be only based on strength of schedule wins/losses. But it is clear that there is also a hyperbolic anti-smurf component which means you can get enormous MMR jumps of something like 2000 points in one game when you start the account. There does also appear to be some in-game stats component that impacts how fast you go up or down.

The whole system is badly distorted at this point because from mid July to mid August the number of active accounts more than doubled. Every new account adds uncertainty to the system until they have lots of games are have a well determined stable MMR. When you more than double the number of accounts that adds so much uncertainty it's almost like you started over and don't know what value anyone should have.

In another couple months it will get more accurate again as all that uncertainty is sorted out...provided the number of players does not continue to grow too quickly. The anti-smurf system is making this problem a lot worse because often new players are adding a bonus (illegitimate) 2k MMR points when they get shot to Very High in 1 game. Even when you average that over 10 players it just screwed up everyone's rating by 200 points when 1 game should only move you +/- 10-20 points after you are stable.

There are a couple things that are in your control to stabilize or destabilize your rating.

[1] Play with far stronger or far weaker players by directly queing with them. If you play with weak players and lose, you can lose a ton of points quickly. Or if you play with strong players and win you can gain a lot quickly.

[2] Play a string of games on your best hero in a row and your rating with jump up a lot. Play a string of games on your worst hero and your rating will drop down just as quickly.

If you want a stable, consistent experience you need to play with a stack of similar skill players and pick only heroes you know. To the extent that you deviate from this pattern you will get matches that don't make sense.

KDA performance

if you scored high KDA, you'll end up on the front page even if you solo queue

WHOSE SMURF ARE YOU??? TELL ME FAST.

who is a smurf?

@kda performance

There are examples of people who abuse their games to get high KDA, and they are also in low skill brackets. Again, chicken-or-egg...

Yeah, there have been several examples of players I noticed who intentionally lost any games that went into High bracket, feeding with a hero they did not like so they could return to easy games and stomp nubs more.

They do tend to get stuck in Low Priority Que this way, but every Sunday I try to make myself not play any games without a 5 stack because those throwers come back out for a day or so....I wish there would be a point where after X number of LPQ trips an account would just be there permanently. I get every report I use back because it's so consistently a thrower on Sunday or Monday from the weekly jailbreak.

Yeah thanks for the reply's guys i have tought about each one of these theory's except the one @S[A]m Fuckin' Peckinpah
said about doing better or worse then the avg stats of each hero.. Btw i opened this thread to try improve my rating

lol @ all the fucking idiots itt writing walls of text

it's an elo rating, dumbshits
go google elo if you dont know how it works

fuck damn you guys are stupid

No, its definitely not an ELO rating system. Although I will grant you that at least one admin on the dev forum erroneously said it was.

yes it is and it's identical to the TMM system, except this one is hidden

I think trying to explain the difference between the behavior of ELO, TSR, and this system is not worth doing since you don't want to know about it and most people will find it boring. Suffice it to say you don't know what the terminology even means.

@Kanye

It's derived from elo, but it's not elo

Elo systems require a 1 v 1 match up, which makes an ELO system impossible in DotA 2. Let me spell it out more expressly for you since you're a moron: DotA is a 5 v 5 game.

Its a factor of things how fast you win games your average gpm per hero how many hero you play well, your win/l ratio, your item builds . Its all based off of an averaging system and if you deviate from the average you will either move up or down

It is an ELO-based system, based on everything I have read and heard on the dev forum.

Simply, you start with X rating and get or lose rating depending on whether you win or lose a game, respectively. You get more rating if the enemies are higher rated than you and less if they are lower rated than you.

In the beginning, they also take into account LH, GPM, EXPM relative to the rest of the team in order to discover whether you are smurfing.

That's not right meow because there is some element of elo in the system. If you lose games to higher rated players and win against lower rated ones your mmr will level off and you won't make very high. Which is the way it's supposed to work, leveling you off so you play competition on the same level and win about half your games.

it's based on ELO but it's garbage, basically you can max out your ELO in your first game on a smurf account and sometimes the MMR is so broken, it doesn't even give a fuck about which skill bracket you belong to and puts you with random low skilled players so you can find a game faster

it depends on winrate....i dont think that it could depend on kda cause what if u re suporting whole game and u got no kills that doesnt mean u re a low skiller

I can probably make a reasonably sized list consisting of players which had low winrates that low winrates and mediocre kdas, but still extremely high hidden mmr.
It simply placed you up against players with equivalent skill level as your own, unless you had extremely high hidden mmr. If this was the case, the mmr system would match you up against/with worse players despite being a much better player yourself.

I see plenty of 48%- wr players in the 3800 area.

Kda, hero damage, stun duration, slow duration and whatever else are used only in first few games, when you can gain more points. After you start losing as much as winning, you gain / lose about 25 mmr per game.

Let's take this guy for example: http://dotabuff.com/players/167468473
Here's the game I had with him: http://dotabuff.com/matches/627226129 (Rexxar)
MMR pic: http://imgur.com/VjoDICJ

You could even find sub 50% winrate players in the 5k+ area ^^

How about 36% players?

Smurf protection goes crazy sometimes!

http://dotabuff.com/players/155740644
Close one. He has 38% winrate and almost 5k tmm with some other guys.

That rexxar bought a lvl 13 account with no games. He probably started at around 5k, and now he is most likely almost back at the same rating he had on his main account :D

You're right! Wave style power leveling! :D Bot matches until 13 then calibration games with tinker.

Edit: OOoo! Bot games on Very High! Nice! http://dotabuff.com/matches/579433488

this is actually my first account http://dotabuff.com/players/140972224, i decided to leave it because i got so many lose from the start so i had to create this one and get good stats before calibration i went for stacks and always go support if needed and have a average KDA on every hero after 150 games i have decided to play RMM and sadly my calibration is only 3800K solo/party, got 64-66% winrate back then and an average of 3.6 KDA through http://www.dotamax.com/.
So i decided to play this account just for fun doing solo queue and randoming heroes but i noticed my winrate is getting higher so i decided to switch account again and yes i only play solo que mostly http://www.dotamax.com/player/detail/140972224/ and get better stats than my previous one... so i decided ill play more normal match making till i have more games in very high skill bracket(got 9 games only at 100% w/ an average of 7.3 KDA -_-) and if more names are seen in a game than anonymous then i can assume im entering higher bracket.
In Conclusion: high winrate and high KDA doesnt really win high calibration in RMM :)

decided to create a smurf one day. got around 40% winrate and 4.5k mmr.
well that didnt work out that great. moved on