21 replies to this topic

[Saturnine]

Normally I don't make suggestion threads at all, but here's yet another.

 

Common complaint loads of people have: Matchmaking is bad and makes people do bad things for bad reasons.

 

How solve?

 

Looking at these forums, many people come to the impression that most of Hawken's players have rather high MMRs - above 1800 is the norm. In fact, players with over 1800 MMR represent a percentage of total players in the single digits. However, there is certainly an overabundance of players at the high end relative to what should technically be possible with the MMR system.

 

These players can get MMRs far higher in proportion to the overall system than should be possible, given the number of players in the system. (Not in the hundreds by the way - Shadeness has ~100,000 unique accounts listed in his system, and goes through several thousand every couple of days.) As a result of this mismatch, these higher level players cannot find matches at their skill level easily. They then have to resort to the matchmaker which takes an exceedingly long time to match them, and then ends up placing them in servers FAR below their skill level, resulting in a massive stomp. Alternatively, they can create smurf accounts so they can have a lower MMR account to play matches with.

 

In order to solve this, I've come up with a really basic system that should draw everyone's MMR closer to a pegged average, while putting a lower ceiling on the extreme ranges of MMR. What should happen under such a system, is players across the board will be more able to find servers that match their skill level, the need to smurf by people at the high end should be heavily reduced, and people should be happier in general.

 

Note that I am NOT a mathematician nor a statistician, so my math is likely pretty bad, or could be improved by someone who is. 

 

Basically, find an average MMR, and artificially peg the average to that number. 1500 makes perfect sense since in my suggestion, there is 1000 points of leeway on either side of 1500. With the pegged MMR average, the average skill level as calculated by the MMR system is set at 1500, forcing the theoretical average player towards an MMR of 1500. At the end of the day with this system, the average should ALWAYS be within a point or two of 1500.

 

Next, set hard limits on MMR at 500 and 2500, instead of 500 and 3000 as they currently stand. Then place a soft limit that is significantly more difficult to advance beyond. Right now a soft limit is kind of in place at ~2850, but I don't think there is a real one. This soft limit should be placed at a floating peg - rather than the hard peg that the average is set at. This floating peg should be placed at the 3.5th standard deviation away from the average. (For an explanation of what standard deviation means: [link]) This should make it much harder for players to advance beyond this point, keeping MMRs from rising into the stratosphere as they do now.

 

Having this soft cap be at a floating point, rather than a hard peg, means that the soft cap will adjust to the playerbase, rather than having a single abusable point. The lower hard limit, in addition to the soft cap means even those who advance beyond the soft cap and hit the ceiling should still have players to match up with, despite being at the very top of the skill level.

 

Here's an example to explain how the soft cap could work. In this example, the soft cap is currently placed at 2300 MMR. A player has 2299 MMR, and just had a moderately good game. His MMR advances one point as a result of his match, making his MMR 2300. For the sake of an example, in her next match, she performs exactly 3 times better as far as the MMR algorithm is concerned. Before reaching the soft cap, she would have gained 3 MMR. Now that she has reached the soft cap however, she only gains 1 MMR. If said player performed poorly, causing a loss in MMR, she would lose MMR at an equivalently increased rate per unit of poor performance, up until she crosses the soft cap line, where she loses MMR at the normal rate.

 

Example: Suicax is at 2305 MMR, and has an unusually poor performance. Since he has an MMR greater than the soft cap, his adjusted MMR loss is 15 points. Instead of losing 15 points straight up, Suicax loses the first 5 points, bringing him to 2300 MMR. The remainder of the point loss is no longer accelerated. He instead loses 3.3 MMR for the remainder of the loss, and finishes at 2296.7 MMR. 

 

An alternate example of the way a soft cap could work is as follows. Apply a flat point value increase in the required score of a given match to gain MMR. If, in a way abstracted example, you were to require 50 points above the server average to gain one point of MMR in a round below the soft cap, above the soft cap, the threshold rises, and instead you need 150 points to gain the same increase. If the base threshold for any increase at all was normally placed at the server average, the threshold is now 100 points above average. If you earn less than 100 points above average while above the soft cap threshold, you instead lose MMR as if Average+100 were the average.

 

For all players who are currently above and below the hard limit, add or subtract from their MMR until they are placed at the hard limit at their respective end of the spectrum. (no one has less than 500 MMR, though.)

 

Here's where the math starts. For all players above 1500, subtract an amount from their MMR equal to (.5(standard deviation * Z-score)). (Z-score is the term for the amount of standard deviations a given number is away from the average.) For a full equation for those above the average :

 

FinalMMR = CurrentMMR - (.5(?*z))          (Where ? = standard deviation, and z = Z-score)

 

Then, do the same thing for MMRs that are below the average, except instead of subtracting from the current MMR, you add to it. The result should be that players farther away from the average are brought closer to average at a sliding rate such that the further away from average you are, the more MMR you lose to get closer to average, while those closest to the average lose comparatively little. 

 

The overall end result of all this should be that overall MMRs across the playerbase are close enough that those at the top can still find servers, and smurfing should be less of a big deal, and everyone overall should have more server options available to them. The addition and subtraction should also be limited enough so as to keep everyone an appropriate distance from the average. (If it is in fact too much pull, just reduce the .5 multiplier by some amount so as to keep a healthy spread.) Having an artificially pegged average will also pull things a bit more towards a nice bell curve, so there will be fewer people way beyond the pale in terms of what MMRs should be possible.

 

Edit: Since I know it will come up - accounts that are unactive for over 3 months should be removed from the process used to calculate the hard average and the floating points. They can be entered back into the system should the account's user return.

 

Again, I am not a mathematician nor a statistician. Take the above with as much salt as you please, but this system should be accurate to what I describe, and I think the end result of applying such a system would be a net benefit for people across the playerbase.

 

If anything is unclear, or if I have missed something, let me know so I can fix it.

 

If my idea is bad and I should feel bad, tell me about it.

 

If you are mad that this means now you'll never make it to 3000 MMR: lolskrub.

 

Edit: Note: I do not think this will resolve all problems with MMR, with smurf accounts, or with high level players having the ability to stomp newbies. Fixing problems like those would require far more stringent measures that would come with negative consequences. I do think that this system will open up the possibility for more players to find more games that are much closer to their skill level. If a player wants to jump into a low level game to stomp them, this won't stop them, just like the current system doesn't. But if you are a 2200+ MMR player, and have trouble finding matches that fit your skill level and give you a fair challenge at present, this system would help you find more of those sweet, sweet, 2 and 3 star servers, more often.

Edited by Saturnine, 10 April 2015 - 09:06 AM.

[crockrocket]

There are a lot of accounts under 500, though I don't know if that is intentional.

[Saturnine]

The accounts under 500 are there because of devs punishing hackers in creative ways. :P

(same with accounts with MMRs over 3000, but I don't think those are there anymore.)

Edit: there was some confusion about my intent with this, so here's a super quick and dirty illustration.

Under my proposed system, the basic goal would be as follows. A player should find it roughly as hard under this system to reach 2333 MMR as it is currently difficult to reach 2800. It's not a perfect illustration, since the closer you are to 1500, it should be almost unnoticeably more difficult to gain MMR. But as you get higher, it gets harder to gain MMR much more quickly than it does at present.

Edited by Saturnine, 07 April 2015 - 08:27 AM.

[Superkamikazee]

It's a plan

[crockrocket]

The accounts under 500 are there because of devs punishing hackers in creative ways. :P

(same with accounts with MMRs over 3000, but I don't think those are there anymore.)
 

Today I learned

[Anichkov3]

I'm not an English speaker, and so it is difficult to perceive so much text. So I want clarify. Do you offer a narrow down range of existing MMR? Seems to be no.

In my opinion you just offer to understate MMR strong players.

My suggestion. If the player MMR is 2400 (or 2500) and above. Just need to fix a lower bound for the server with 1 star - is 1900 (or 2000). And the upper limit for the server with 1 star - is 3000+.

Edited by Anichkov3, 07 April 2015 - 08:57 AM.

[CrimsonKaim]

But what happens when you pull higher tier players together at an average point wit haverage and low tier players? Wouldn't that actually lead to MORE stomping?

 

Because I understand the whole thing as follows:

 

1. Pull everyone close to an average point.

2. Take or add MMR points to higher and lower tier players so they don't get too far away from it.

 

Result: A bunch of palyers randomly mixed into matches.

[Saturnine]

There would be a little bit of mixing of higher and lower, but not too much. Those at the highest end would still be at the highest end after adjustments. They would be more about to join lower servers, which could be avoided by restricting the MMR range of access, but you'd have to be careful with that, otherwise you'd risk defeating the entire purpose of this change.

Players overall will be closer together, but I don't think you'd see people at the high end dipping very far into the low end. At the very very least you'd see less of it than you do presently.

[CrimsonKaim]

But how should that work? On the one hand you wnat to have more players from different skill levels to paly together (this is not represented by MMR), on the other hand, you want in this more random mxiture of players to have more balanced games by only putting players of their skill level in their skill level matches?

 

If you pull players closer to a certain point, you will automatically support the stomping we have now, on the other side, adding mroe restrictions will force me to make smurf accounts. 

 

Ofcourse a bigger player base will do but I think MMR should not be taken into account that much, more stats as suggested by another guy in another thread. In TDM, look at KDR. In Siege and MA, look at Team Win percentage. In DM, you take the MMR.

While not leaving other factors out, these mentioned above should be the main priority. 

 

A Technician with 1/14/28 stats is still a bad Technician.

 

 

EDIT: Oh, and get rid of MMR restrictions. Atleast make them wider, in the current version it denies me to join any game, even though I intentionally lowered my MMR and manipulated the MM. So everyone is able to play.

Edited by FakeName, 07 April 2015 - 09:34 AM.

[Anichkov3]

No. MMR range should extend only to the players have reached a "critical point" - over 2400...2500+.  

You know that the server without the stars for you is - if your MMR over +-500 different from the average rating of the server.

[Saturnine]

EDIT: Oh, and get rid of MMR restrictions. Atleast make them wider, in the current version it denies me to join any game, even though I intentionally lowered my MMR and manipulated the MM. So everyone is able to play.

 

This problem is exactly what I'm trying to fix with my idea. I contract the overall variance between MMRs so that players like you, who have to intentionally lower MMR to get into a game, can actually find games to get into. The way I suggest doing it would contract the high levels of MMR so the people at the top can fight against people their own skill level more often, and without having as much impetus to create a smurf that has lower MMR so they can find games to play. I suggest it like this so that people have more options, but not make it easier for people to get into servers they don't belong in.

[CrimsonKaim]

But how do you adress the problem with more players in one room stomping pubs?

[Saturnine]

But how do you adress the problem with more players in one room stomping pubs?

A big part of why high level players are stomping low level servers is because those are the matches the matchmaker puts them into, since there aren't any matches that are their level available for the matchmaker to place them into. By condensing MMRs, there will be more servers available for more people to get into, since high level players won't be spread out across 2000-2900 MMR, and will instead be mostly between 2000 and 2300, which should allow more high level players to get into the same servers, rather than using the matchmaker and hoping it doesn't drop them too low, or instead of creating a smurf and smashing their way up from the bottom over and over again.

 

Edit: VVV The one aspect this won't really help is people deliberately creating smurfs or joining low level games with the intent to stomp all over them. Unfortunately math cannot quite correct for the human desire to feel powerful by exerting their power over people with less power than they do. 

Edited by Saturnine, 07 April 2015 - 02:28 PM.

[CrimsonKaim]

Hm ... it does not solve the actual problem I g uess. by lwoering the overall distance between MMR "borders", you do not neccessarily change the actual skill players have.

 

I would do this with some caution, but definately worth a try.

[Jerv]

Personally, I've always been a percentile guy. It seems to me that if you took the relative rankings of all current players instead of an absolute rating and used that instead, then a few milliseconds of calculation could prevent many hours of queue waits.

 

Basically, it's a sliding scale with "stops" at the top and bottom end so that the best and worst players will have the same number of potential allies/opponents as those players closer to the median. In other words, instead of putting only those with an MMR between X and Y into a match, it would say something more like, "Of the 1,000 players online right now, you're the 287th best, so I'll pick a few people between 187th and 387th and put y'all on the same server.".

 

Here's the rub; for purposes of that calculation, your effective ranking will never be closer to the top/bottom than half the range; that example above has a 200-point spread, so the top 100 players will all have the same effective ranking (as will the bottom 100) and thus the player pool that the MM chooses from will ALWAYS be 200 people no matter what the MMR score.

[Saturnine]

I've added an example of how a soft cap could work in such a system to the OP, as well as here!

 

Here's an example to explain how the soft cap could work. In this example, the soft cap is currently placed at 2300 MMR. A player has 2299 MMR, and just had a moderately good game. His MMR advances one point as a result of his match, making his MMR 2300. For the sake of an example, in her next match, she performs exactly 3 times better as far as the MMR algorithm is concerned. Before reaching the soft cap, she would have gained 3 MMR. Now that she has reached the soft cap however, she only gains 1 MMR. If said player performed poorly, causing a loss in MMR, she would lose MMR at an equivalently increased rate per unit of poor performance, up until she crosses the soft cap line, where she loses MMR at the normal rate.

 

Example: Suicax is at 2305 MMR, and has an unusually poor performance. Since he has an MMR greater than the soft cap, his adjusted MMR loss is 15 points. Instead of losing 15 points straight up, Suicax loses the first 5 points, bringing him to 2300 MMR. The remainder of the point loss is no longer accelerated. He instead loses 3.3 MMR for the remainder of the loss, and finishes at 2296.7 MMR. 

 

An alternate example of the way a soft cap could work is as follows. Apply a flat point value increase in the required score of a given match to gain MMR. If, in a way abstracted example, you were to require 50 points above the server average to gain one point of MMR in a round below the soft cap, above the soft cap, the threshold rises, and instead you need 150 points to gain the same increase. If the base threshold for any increase at all was normally placed at the server average, the threshold is now 100 points above average. If you earn less than 100 points above average while above the soft cap threshold, you instead lose MMR as if Average+100 were the average.

Edited by Saturnine, 09 April 2015 - 06:23 AM.

[CrimsonKaim]

I see alot of players have an MMR between 2299 and 2301 ... :/

[SigmaOmega]

You have my vote. I would like to have a mix of average and little more skilled players in a match

[bluebaron201]

When you say pull players towards 1500 MMR are you talking about an MMR decay, a stronger dampening factor, or something else entirely?

Edited by bluebaron201, 09 April 2015 - 12:15 PM.

[Saturnine]

I see alot of players have an MMR between 2299 and 2301 ... :/

 

That's why the soft cap is floating based on the 3.5th standard deviation, and isn't a fixed point. I used 2300 as an example, not where the soft cap should be, at least not at all times.

 

 

When you say pull players towards 1500 MMR are you talking about an MMR decay, a stronger dampening factor, or something else entirely?

 

 

The idea would be that with an enforced average MMR of 1500, as the distribution of MMRs change, peoples' MMR will change based on which players have however much MMR, in whatever distribution. As people earn whatever MMR they end up with, MMRs across the population will increase or decrease by whatever amount necessary to bring the mean MMR back to 1500, rather than, say, an average of 1502. Since the population is in the several thousands, the adjustment would be extremely small and be hardly noticeable unless you were watching extremely carefully for it.

Edited by Saturnine, 09 April 2015 - 03:53 PM.

[JackVandal]

i read this, understood it, but could not explain that without reference, i think its a good idea but it would have to be implemented right. but something does

need to be fixed with the current system and the way it changes, iv had MMR vary by 200 over 3 days at some points.

[AsianJoyKiller]

As long as the matchmaker can be completely circumvented, there's always going to be issues with MMR.