Forged Alliance Forever

Re: new rating system & diversification ideas for maps

2018-11-08T23:08:36+02:00

This would improve the overall accuracy of rating for sure. But it would be also nice for the rating system to encourage people to diversify their map palette
It would be a nice compact way of delivering both carrot and stick to people Thats why i posted my sweats above to clear things up

And i dont think it would be very invasive cos still people would be able to play all the maps they want, but it would also convince some of the community to broaden their spectrum just for the sake of points or "whatever, why not, i can only gain from this"

Statistics: Posted by Ajack — 08 Nov 2018, 23:08

Re: new rating system & diversification ideas for maps

2018-11-08T22:54:57+02:00

fwiw, I worked out the technical details of how to adapt trueskill to track rating as a function of map-class (as well as a few other things) a couple years ago. I put the results here: /viewtopic. ... 8&start=10. Short answer is that just dividing maps into small, medium, large classes allows trueskill to almost halve its uncertainty about a players predicted performance for any given game.

Statistics: Posted by Axle — 08 Nov 2018, 22:54

Re: new rating system & diversification ideas for maps

2018-11-08T23:34:56+02:00

I dont know if what i'm saying now was entirely covered before cos i must admit to the crime of "skimming" some of previous posts xd Pls dont hate me

BUT, i always belive that everything must be simple and idiot proof so there shouldnt be right out supercomplicated systems cos everybody will just say "f... that"

What if (and again plz dont roast me hard for being stupid and not knowing things) the rating system had somthing like that:

gain of points == (fixed ammount) * (player count modifier to the power of game type [FFA << 0,5 >>, Teams << 1,5 >> etc]) * (game balance modifier [i think its in?]) *
( <<magic map modifier>> ) * ( <<magic anti stack modifier>> )

loss of points == (fixed ammount) * (player count modifier to the power of game type [FFA << 0,5 >>, Teams << x 1,5 >> etc]) * (game balance modifier [i think its in?]) *
( <<magic map modifier>> to the power of -1 ) * ( <<magic anti stack modifier>> to the power of -1 ) { this one may be dangerous to implement}

so the magic map modifier would start with [0,9] and it would degrade to [0,3] when repetitively playing the same map and it would grow back to [0,9] when playing different maps

magic anti stack modifier would also degrade from [0,9] to [0,3] when repetitively playing with the same team composition to prevent friend stat padding

( the stack part i think would be hard cos it would need a lot of data storage but maybe not i dunno, i'm a bad programmer, programming only factory software for machines )

Soooo... that way i think... every time someone plays the same map over and over he gets less points for winning and looses more points for loosing. Basically benefits would wear out and penalties would grow every time you play the same map to the point of loosing max points and not gaining any.
I'm not sure about that part with loosing points cos it would make grounds for smurfs to degrade their stats quickly probably.

Also my magic modifiers would be reset once a month or week to not be so harsh. But they could be permanent xD

The way at which both modifiers degrade or grow would also be dependant on few more factors but that would need more consideration

Statistics: Posted by Ajack — 08 Nov 2018, 22:29

Re: new rating system & diversification ideas for maps

2017-12-07T15:37:33+02:00

CookieNoob wrote:
I dont think a matchmaker would solve the issue

How?

Statistics: Posted by TheKoopa — 07 Dec 2017, 15:37

Re: new rating system & diversification ideas for maps

2017-12-07T14:24:45+02:00

Great idea! This will help with a lot of issues.

Maybe a linear weight isn‘t the way to go though, but that‘s just details.

Statistics: Posted by SpoCk0nd0pe — 07 Dec 2017, 14:24

Re: new rating system & diversification ideas for maps

2017-12-07T14:10:04+02:00

sigh.

since i got quoted out of context id just like to point out that:

1. me being pinged by people complaining about their rating changing because its shown exactly and not rounded, isnt relevant to this argument. quoting me on this doesnt help your suggestion. thanks. neither does your suggestion solve this particular issue in any way.
2. this only illustrates that understanding rating is already a problem. your suggestion will only make it worse. the argument that "its already bad therefore it being worse isnt an issue." is not good. thats exactly how you end up with a complete mess.
3. theres an order several orders of magnitude of difference between me being pinged and actually having to categorize, maintain and update all the map vaults maps in a table for rating systems, fixing all the bugs that come with this, handling any edge cases, updating the server, all for something that is subjective anyway. this comparison is ridiculous.
4. its extra funny because my issue will actually become a lot worse as well - now ill be pinged even more by people asking me why their rating suddenly dropped 400 points when they joined a wonder and not a thermo.
5. not even mentioning the part where trueskill is data driven so doing this would actually f*** with it in ways no one here has made actual, evidence based predictions for

as brutus said, matchmaker would take less time than this whole mess, and while its not the exact same issue, it will solve a much bigger set of problems for faf, rather than a select few (high rated players who consider themselves to know best) complaining on the forum about something thats at best arguably good.

i would like to pay attention to priorities please. this discussion is great and all but i honestly dont expect anyone to actually start work on it. way too much effort for little gain.

Statistics: Posted by Exotic_Retard — 07 Dec 2017, 14:10

Re: new rating system & diversification ideas for maps

2017-12-07T13:29:57+02:00

Brutus5000 wrote:
I am missing a few aspects on this:
How will these multi-dimensional ratings be calculated? What are the base values for calculating each dimensions new rating? If I understand it correctly, each rating should be calculated independently from each other and only the final rating should be a mix of these values.

Yes in this suggestion, the ratings are propagated separately. I.e. the rating after the game is only calculated inside each affected category. The rating that is used for balancing the game is a combination of the relevant map tag / player count / global rating

Brutus5000 wrote:
There problem addressed here seems to be, that you want to downrate players, that gain rating on only one map.
If you calculate them independently the proposed changes will backfire in the opposite direction: low-rated players will be uprated on maps they play for the first time. Since the initial rating is pretty high, the rating in the 600-1000 range for players with few games becomes even more volatile, killing the little rest of balance you have in that range.

Trueskill adjusts within a few games to the most probable rating. We can use the already played games to calculate the rating for each category for each player (we dont have to start from scratch). Only very few players will start in this system with the 1500/500 mean/deviation for their categories. Even if some categories are on these values, this only contributes to a part of their rating and thus the volatility is not that big.

Brutus5000 wrote:
From a support perspective these changes are a nightmare. Even today many player have problems understanding why the rating changed the way it did. And there are certain aspects - like joining a new game before the old one ended with ratings override each other - that make it even more confusing. Making the rating logic-multidimensional will make these problems even worse.

Even today there are problems with this (quoting a mod who has to deal with this daily):

Exotic_Retard wrote:
point is, is like literally 20 people in faf understand how rating actually works

Yes it is a system that is not really easy to understand, but (as it was said before) the players dont really have to calculate it themselves. Solutions could be to show in the game lobby "Your estimated rating for this game configuration is approximately:" or to hide rating completely and replace it with a level system (i.e. bronze, silver, gold,...) depending on the calculated rating for the game. This means that a player could be gold in one game and silver in another one.

Brutus5000 wrote:
From a developer perspective I'd estimate, that implementing a matchmaker should be less effort than implementing this change. But anyway: nobody is willing to touch the current server code. So we'd have to wait for the java server, which is waiting on ICE, which is in development for more than a year with still no light at the end of the tunnel.

I dont think a matchmaker would solve the issue, but I also know that it will take a while until this system would get implemented if it is decided that we want it

Statistics: Posted by CookieNoob — 07 Dec 2017, 13:29

Re: new rating system & diversification ideas for maps

2017-12-07T02:18:52+02:00

Brutus5000 wrote:
I am missing a few aspects on this:
How will these multi-dimensional ratings be calculated? What are the base values for calculating each dimensions new rating? If I understand it correctly, each rating should be calculated independently from each other and only the final rating should be a mix of these values.

There problem addressed here seems to be, that you want to downrate players, that gain rating on only one map.
If you calculate them independently the proposed changes will backfire in the opposite direction: low-rated players will be uprated on maps they play for the first time. Since the initial rating is pretty high, the rating in the 600-1000 range for players with few games becomes even more volatile, killing the little rest of balance you have in that range.

From a support perspective these changes are a nightmare. Even today many player have problems understanding why the rating changed the way it did. And there are certain aspects - like joining a new game before the old one ended with ratings override each other - that make it even more confusing. Making the rating logic-multidimensional will make these problems even worse.

From a developer perspective I'd estimate, that implementing a matchmaker should be less effort than implementing this change. But anyway: nobody is willing to touch the current server code. So we'd have to wait for the java server, which is waiting on ICE, which is in development for more than a year with still no light at the end of the tunnel.

I can't speak for others, but I'd rather have divisions for gameplay. Like 1v1, then 2v2/3v3, then 4v4+ maybe. In my eyes these are the three situations where gameplay is different enough to require varied ratings. Divisions could be determined by rating that is shown now with certain cutoffs at some ratings. The current system is pretty illogical in the way that it assumes rating difference is equal at all rating ranges. A 1900 is going to play an order of magnitude better than a 1600. A 1000 beating a 700 is more of a flip of a coin, really. Divisions would make balancing games out a little bit better on that front. I guess the problem would be having different ratings for these game modes and then being in a lobby that switches from 4v4 to 2v2.

Personally I would rather that you gain progressively less rating per game you play on the same map. I just want that change because I feel it could at least subtly push people to branch out rather than farming a single map. Of course the same objective could be accomplished by some sort of "featured map" idea as was discussed by some others. Or it could be a combination of the two.

Statistics: Posted by FtXCommando — 07 Dec 2017, 02:18

Re: new rating system & diversification ideas for maps

2017-12-07T01:50:05+02:00

I am missing a few aspects on this:
How will these multi-dimensional ratings be calculated? What are the base values for calculating each dimensions new rating? If I understand it correctly, each rating should be calculated independently from each other and only the final rating should be a mix of these values.

There problem addressed here seems to be, that you want to downrate players, that gain rating on only one map.
If you calculate them independently the proposed changes will backfire in the opposite direction: low-rated players will be uprated on maps they play for the first time. Since the initial rating is pretty high, the rating in the 600-1000 range for players with few games becomes even more volatile, killing the little rest of balance you have in that range.

From a support perspective these changes are a nightmare. Even today many player have problems understanding why the rating changed the way it did. And there are certain aspects - like joining a new game before the old one ended with ratings override each other - that make it even more confusing. Making the rating logic-multidimensional will make these problems even worse.

From a developer perspective I'd estimate, that implementing a matchmaker should be less effort than implementing this change. But anyway: nobody is willing to touch the current server code. So we'd have to wait for the java server, which is waiting on ICE, which is in development for more than a year with still no light at the end of the tunnel.

Statistics: Posted by Brutus5000 — 07 Dec 2017, 01:50

Re: new rating system & diversification ideas for maps

2017-12-07T00:25:24+02:00

Nobody is thinking about DIScouraging Gap or Thermo or any other map. They are thinking about ENcouraging maps other than those. It's a different mentality. The former insinuates a desire for control and oppression, the latter a desire to allow people to choose to branch out. It's a simple fact that one reason people stay doing what they do, in any aspect of life not just supcom, is because they have never been exposed to another option.

Statistics: Posted by IceDreamer — 07 Dec 2017, 00:25

Re: new rating system & diversification ideas for maps

2017-12-06T23:52:18+02:00

well that's the alternative.
But then you can become ""good"" on gap, and be under-rated when you start your first "out of gap" games.

there is drawback and advantages in both.
I guess cookienoob way of doing this is just making sure what i said before doesn't happen. But they are quite similar in the end.
And tbh you don't need to care about how your rating is calculated, if it's complicated or not. Algorithm can do it for you, really.

Statistics: Posted by keyser — 06 Dec 2017, 23:52

Re: new rating system & diversification ideas for maps

2017-12-06T23:16:58+02:00

Wouldn't it be easier to leave the global rating as is and have a "featured maps rating" (a real team-game ranking) set apart for maps we consider to be "the real supcom experience".

I came to this idea while thinking to unrank gap/thermo games, but instead of generating a fuss by discriminating them, we can discriminate the good selected (aka featured) maps.

I don't think we need a system that rates you based on the number of players (2v2, 3v3...) in the game. That's unnecessary complicated, really.

Statistics: Posted by angus000 — 06 Dec 2017, 23:16

Re: new rating system & diversification ideas for maps

2017-12-06T22:50:23+02:00

CookieNoob wrote:
An example would be a player that plays 100 games on gap only and has a rating of 2k there. That player has a rating of 2k (mean rating - 3* deviation) in the global rating and in the map category "gap" and in the "4v4 game mode". When he now joins another 4v4 game on a different map with other tags, he would have a rating of 30% x 2000 + 50% x 0 (map tags are not played before, so he has 1500 mean - 3x 500 deviation, this value will adjust very fast due to the high uncertainty. Approximately after 10-15 games the rating will already be in the correct range) +30% x 2000 ("4v4 game" game mode rating) = 1000 rating. If the game is 1v1 then he would have a rating of 30% * 2000 + 50%* 0 + 30% * 0 = 600 (this is not accurate at first and will take a few games to be more precise ofc). Keep in mind that these numbers are just arbitrary suggestions.

and if you are in a gap lobby with him wouldn't it show just 2000?

Statistics: Posted by nine2 — 06 Dec 2017, 22:50

Re: new rating system & diversification ideas for maps

2017-12-06T22:12:42+02:00

?

Farms, ftx, econoob and Aleg on a team together lost to the king in hand to hand combat on dual gap.

Statistics: Posted by Feather — 06 Dec 2017, 22:12

Re: new rating system & diversification ideas for maps

2017-12-06T20:09:27+02:00

PhilipJFry wrote:
exact values are up for debate
yes a map vault maintainer/curator would be needed to ensure the system works
even if you play gap only once every full moon it will be ok since the massive initial deviation allows for rapid changes to the rating

gaining less rating from playing the same map over and over would just make it easy to lose rating on custom 1v1 and then "smurfing" on your map of choice

imo we can choose 3 paths:
1) improving global rating to the best of our ability to allow people playing their map of choice while being rated
2) make custom games unrated (and hopefully get a auto match maker for team games)
Spoiler: show
3) nothing

I know the values are up for debate that's why I wanted to reverse the quantities. I don't really see that being a problem with a system where 1v1/2v2/3v3 etc have varied ratings. Really if you wanted to smurf 6v6 dual gap you would need to purposefully lose on other 6v6 maps. Otherwise you would be affecting your "1v1" division or something. Not really sure whether what I'm saying is practical or not.

A "Vault Councillor" would be able to stop the abuse of losing on 1 slightly different version of dual gap to farm on another version of dual gap, hopefully.

Of course I always would support a team matchmaker solution. It makes all this completely unnecessary. However until that gets into the work the best we can do is improve the global rating situation.

Statistics: Posted by FtXCommando — 06 Dec 2017, 20:09