Once you start doing it, you can fine tune it.
It doesn’t need to be exact every time. It just needs to push your MMR in the right direction in overall sense. We do that everywhere with various engineering solutions, they are rarely exact. Weather forecasts can be pretty awful at times, but in overall sense they are very helpful and we heavily rely on them.
2 Likes
here you go:
1 Like
I get the analogy to weather forecasts, but you also have to realize that it is a completely different situation. An algorithm has to be consistent. We have models to predict weather based on usual patterns. Weather isn’t intelligent. If you could find any way at all to play terribly, but still raise your MMR, people would exploit the sh*t out of it. Blizz has to take all of that into account.
Lol
Thanks for the replies everyone 
i agree that it wouldn’t be easy to do and testing for sure needs to happen. In the current state most are saying the game is on “life support”. I guess time will tell what happens.
The way the system works is:
It keeps track of every single game.
It compares every element in every game.
It checks which elements were higher in games won, and makes them the factor that adjusts your MMR.
So if a hero wins games consistently by taking average of 5 merc camps per game, while also dealing at least 2000 damage per minute of game time, players who take 5+ merc camps per game (with that hero) and deal at least 2000 damage per minute of the game will get bonus MMR. However if one takes 10 camps in a game, but has less than 500 damage per minute of game time, he’ll lose MMR.
1 Like
What if your team wins and everyone on either team has low damage based on average in that mmr. What if getting 10 camps was what was necessary to win that game? What aspect helped him get the 10 camps? Did he have time because his team had a lead and he didn’t have to worry about xp as much? There are tons of other variables to consider there and about half of them are tied into how well off you are doing in the game which is tied to a shared experience pool.
Alan Dabiri and Mateu Cooper are a little stupid(if you know who they are…) they…
MUST THINK ABOUT TO INCREASE POINTS BASED OF THESE 20 ACHIEVEMETS FOR WIN GAME AND 200 POINTS WOULD BE SAME FOR LOSES GAMES.
THESE 20 ACHIEVEMENTS SHOULD BE GARDITIONE BY IMPORTANCE START FROM MVP SO…ON SO IN…AND POINTS WOULD BE DECREASED BY " -10" PER LEAGUE FROM BRONZE TO GRAND MASTER…WITH THIS WAY…CAN BE CLIMP UP A LITTLE BETTER AND OFFSET LOSES FROM BAD PLAYERS AND LUCK SITUACIONS AND THIS ALMOST 50 % WIN RATE.!!!
POINTS WILL BE MAX 300 FOR MVP AND BETWEEN 20 POINTS BASED OF THESE 20 ACHIEVEMENTS AND START BE LOWER PER EVERY LEAGUE.THESE 200 LOSES FOR EVERY LOSES GAMES POINTS CAN BE OFFSET WITH THESE INCRESED WIN RANKED POINTS.
FROM THESE 20 ACHIEVEMENTS 6 ARE MUCH MORE IMPORTANT!.IF ONE HIGH GAME XP-ED AND KNOW HOW DO NOT DIE OVER 5 TIMES…AND DO OFTEN SOME OF THEAE 6 IMPORTANT ACHIEVEMENTS…CAN A LITTLE CLIMP UP THROUTH THESE 5 DEVISIONS!!!
BUT RIGHT THEY ARE STUPID BOTH LEADER GAME DESIGNERS (MATU AND ALAN)…
HOW MANY YEARS ARE NEEDED FOR THEM TO THINK ABOUT THAT?!?! A HOW MANY?!?!
JESUS…
um…i cant make any sense out of what you just said.
achievements have been in the game for years and give portraits. the event quest system has also been going on for years and they have been doing various things like mounts, banners, portraits, extra gold, etc.
In practice performance base MMR is too complex to write algorithms for.
Humans can barely agree on who the better player were.
Also w/ the player base so small, it really would throw things off. I still think they could of come up with something to bring the outliers down/up faster. i.e. if you are dying 2x more than similar role/rank/toon, you drop like a rock.
This topic must be pin in 1st page of forum!
1 Like
It’s amazing how people are so gullible to think a system like personal based matchmaking rating is possible.
I get some people don’t understand the scope of all the variables to judge what makes a successful game, and whittles them down for it should be easy. The fact that Blizzard had the hubris and arrogance to think they could just baffles me to this day.
I see self driving cars a reality as hard as that might be. However, those are still relatively fixed variables. You have a destination, you have strict paths you have to follow, you have scanners that can detect incoming objects and how fast you’re going to adjust.
Now imagine, if your car could die, and could ressurect, and you could spec it for different talent paths, and you were co driving with other cars, but each of those cars had a different function for the road, and you’re also going against traffic with other cars who want to kill you. Then you have pedestrians who you have to knock over on the other side, but you’re also hiding behind your own, and there isn’t a set path, there’s a destination, but there’s multiple ways to get there and even if you get there, it also starts shooting at you.
Figure out the best way to see how to win that, and oh yeah the game state changes with every patch, and every player. Oh, and Blizzard isn’t several different industries devoting billions of dollars to figure it out.
I think the most embarrassing thing is they think they could actually do it in a respectable and accurate way, and then sold the naive public that it was possible.
2 Likes
I don’t think it was hubris to try and develop a system that might help with judging play to adjust matchmaking. I’ve no doubt they understood at the time that it was an incredibly difficult thing to achieve. The community had been demanding such a thing since the beginning. They may not have called it a PBMM system, but people have wanted to be judged on their performance and not ‘dragged down’ by their teammates.
It is the wisdom of Blizzard to try to implement a system that would help and then pull back when they realized it was not functioning the way they wanted it. Maybe they should have never spent resources chasing that idea but if no one tries to push the envelope, then we stagnate.
I Just want to bring up one of my own posts from some weeks ago, talking exactly about the PBMM.
And to answer to your post: Yes, maybe the PBMM system didn’t work properly on GrandMaster levels, since there it takes really a lot more than having good numbers in a game to actally be good, but for lower leagues it could have made a big difference. Up to gold and probably even diamond it’s quite easy to tell a “good” player from a “bad” one by stats alone - a tank that always dives into the enemy team and dies the whole time, since he used his escape to go in; a healer that is sololaning with a lot of xp contribution but barly any heal; a specialist with a lot of siegedmg but no hero dmg, since he never came to any objectives; an assassin with less dmg than a tank, since he never hit any skillshots; and 1st of all - a player with lots of death, because ppl don’t understand that dieing is the worst thing in the game to do, even if you are a Leoric, Tyrael or Uther, that has some “trait value”. HotSlogs works with those numbers too, and I usally find it quite adequat; only healers and tanks shouldn’t be expected to do siegedmg.
And it was always said by the devs, that the PBMM would only be introduced to leagues up to Diamond, if I remember correctly.
3 Likes
The way they wanted to implement it sounds like a huge “minority report” central hub tbh, I mean = have an AI that learns good play over iterations of games and then compares to your actual play ?, might’ve worked but srs doubt so
Such a system is
A = too expensive (and complex) to maintain
B = too much variables to implement
C = too much time for it to “self-learn” patterns and stability
And yes, that’s quite a bit almost too ambitious and the biggest question of all is how would it detect “personal stat padding” vs “actual team impact” in a game ?, call it whatever you want but sounds far too ambitious to have worked
Still better do/change something (as opposed to nothing) cause probably nothing can be worse than current situation of “evaluating skill” by solely/exclusively? WR% though
1 Like
hmm…you dude…look so you are so smart than Blizzard team… none from them didnt think about that befor. I like so much your suggestione. I hope so Blizzard read your reply and work on it such increased ranked points would so better than now current versione. !
Why couldn’t they have at least started small though?
Like get these 12 death kael-thas guys with less damage done than tanks off my team for love of thralls balls.
2 Likes
SJW made Overwatch… SJW ruined OW and…they removed downvotes…their report system gave the “death by a thousand cuts” fate. They already did the damage.
Which is why machines need to do it. They’re better at seeing certain kinds of patterns in large volumes of data.
But to truly do it they’d really have to work with Google’s DeepMind like they were doing with Starcraft.
It would need to watch replays and not only consider wins/losses, but consider what kinds of actions lead to more sucessful team fights, or quicker experience gains, or more map control.
Using Nova as an example.
It’s easy to just look at damage, or kills/deaths, or how many snipes are hits/misses, or even how often snipe sits off cooldown with Nova not actually contributing.
What isn’t easy is looking at her clone usage.
When she drops her clone in a high traffic area as the team is approaching the objective, reveals an ambush, and an ally turns around in time to avoid being blown up, how many things do you have to consider to give that FAR more weight than her just dropping it randomly? Because that simple play is a big deal. If that ally gets bursted down before the team fight even starts, what are the chances the rest of the team will rush in and die in a 4v5? At low levels, WAAAY too high. Which then snowballs toward defeat, she essentially just saved an entire team fight, and possibly the game. Or at least prevented a lot of uphill fighting.
So it’s very much a difficult undertaking.
And even if it were possible, the AI would have to analyze each game to look at those cause and effect situations before giving you the adjustment results. Because it’s apparent that mere stats are both abusable, and not actually all that telling of skill, so they couldn’t just generate numbers and just use those.
How long do you think that would take? You wouldn’t know the impact of a game until quite a while after it was actually played, MMR adjustment would come at the end of every season, instead of at the end of every game.
I mean I’m all for it if they can do it.
And if they want to just shortcut it and make it about stats, get the double digit death people off my team forever, I’d be fine with that as a compromise. Just stick me with people who have similar stats with each character. Who acknowledge that I play Valeera differently and don’t have high hero damage but do a lot of tanking, peeling, and set up.
If you think a computer can say who was better/worse in a match more accurately than a human is capable of doing, you are probably bad at the game or do not understand the limitations of computers.
Its too complex and it would be easily gamed by players.
If you think it can’t, you probably don’t see the direction AI is going.
I didn’t say there was one capable of doing it now.