Ranking overhaul/reset

[Lizard King]

How hard would it be to go further with the ranking system? I'm talkng similar to marius or playmyth where losses incur point loss, ffa wins garner more points, point values awarded are weighted by the ranks of ppl in game (i.e. beating the comet while you're two daggers would give you significant points while they'd incur serious losses).

A ranked reset has also been proven to reinvigorate the game for a bit during a lull period, at the very least i think it would be good for the community, thanks.

-All Time FFA Champ

[Asmodian]

I haven't been playing, but I agree with LK. A rank reset and some minor changes to the scoring would increase activity slightly.

[punkUser]

Rank reset is pretty easy, as is modifying the score formula somewhat. It's more work to do some sort of ELO-style system where rank gain/loss is based on relative opponent ranks and I'll repeat that it's fairly low priority right now since it simply is not possible to do any reasonable ranking ladder with so few players. Since folks don't generally believe me on this one (despite having tried to explain it a few times), here's another take from a different developer:
http://joostdevblog.blogspot.ca/2014/11 ... rmous.html
Note that while he's speaking about "matchmaking" there the problem is actually equivalent or worse for ranking in general, as the point of matchmaking is to generate matches with as useful information content in the results as possible. With poor or no matchmaking (due to low player counts or similar), there is simply not enough useful information in the outcome of a game to base rank on... you end up measuring noise and the ability of the players to exploit the algorithm MariusNet keeping the algorithm a "secret" is actually indicative of the fact that it was a problem then too, even with far more players.

So yeah I can easily reset things and I'm happy to modify the scoring for a given game - including negative scores - if people can agree on how they want it to work. If you guys recall the current scoring was arrived at via a desire to encourage larger 2-team games with more players rather than just cycling 1v1s/ffa quickly and indeed it was successful for a while. Happy to do something else, but it has to be based on a single game result in isolation, not metadata about the "rank/scores" of other players in the game.

Let me know what sort of scoring you'd like to see. Ideally pick some example games from http://gateofstorms.net/tournaments/RS1/rounds/351 and walk through how they should be scored for the proposed algorithm. Ex. it's totally possible to use something like Milk's FFA scoring for the ranked scoring as well for instance, and having a season focused on one game type is probably not a bad idea to start with given the low player counts.

Cheers.

[Lizard King]

If it can be done easily, copy playmyth's ranking system. I'd have to guess someone still has their hands on those algorithms.

The current system just rewards whoever plays the most games.

[Lord---Scary Owl]

Its quite annoying to see terrible camping FFA players with moons and novas that have negative ratios and are terrible at 1v1. Just because they play KOTH PG over and over doesnt mean they are better. I've seen lots of people with higher symbols than me that I can crush on a daily basis. Ex: dbnova, malsleizure, jojojunglepig, edulus, vihaan, JI. Even though people have high ranks doesnt mean they are good. Is there some sort of system that can correct that and make differences of points based on differences of games? Ex: FFA Win ~ 20 Points, 2T Win ~ 30 Points, 1v1 Win ~ 40 Points. You could still climb your way up while also showing dominance in the 1v1 field. Is this system possible?

[Lizard King]

Rank has almost never been an indicator of skill really. A 1v1 win should never be worth as much as a ffa win, weighted games based on ranks would be nice; basically just copying playmyth/bungie's ranking system.

[punkUser]

If it can be done easily, copy playmyth's ranking system. I'd have to guess someone still has their hands on those algorithms.

Last time I'm going to say this - all those systems are just basic Elo and they do not work without lots more players. There's no debate on that, it's just math and I'm getting tired of explaining it. Even the most basic ladders work with lots of players, it has nothing to do with it being a special "algorithm".

The current system just rewards whoever plays the most games.

Exactly, by design. Don't you remember the entire discussion around this originally? It is designed precisely to encourage people to play more, larger games.

Anyways I'm still open to different scoring systems but like I said, not Elo-style unless we suddenly get a few hundred active players minimum. We don't need to retread this ground every time... we were all on the same page when we implemented that last ranking system and there's really no new information.

So throw out scoring algorithms for specific games (remember: must work in isolation for one game), or else I'll just do a reset with the current system. Or switch to FFA league scoring or something for fun.

[Lizard King]

Seems like it would work just fine to me. There is no specific reward to the current system other than playing more than others will garner a higher rank, it goes against the concept of ranking players in and of itself. You don't even lose points for losing a game nor do I see how this could possibly correlate to "creating more larger games", take it from someone with years of experience.

I guess if we're just going to keep the same old crap reset it at the very least...

[punkUser]

Seems like it would work just fine to me.

Yeah well it turns out that math is more useful than feelings in this instance of... math I hate to be the guy saying that I probably understand this more than other folks here, but I probably understand this a lot more than other folks here. You're going to have to trust me unless you've done some detailed reading on how the mechanics of ranking systems works.

There is no specific reward to the current system other than playing more than others will garner a higher rank, it goes against the concept of ranking players in and of itself.

Right, although you can obviously increase in rank more quickly by performing better. But that's fundamentally the point so I will repeat it - there is a requirement of a minimum player base to establish meaningful rankings in terms of predicting who will win against who, and that requirement is even higher without matchmaking. Nothing predictive can be established with so few players and games so you have to abandon that idea to start with.

Thus if we're going to be playing a "points metagame" rather than "ranking", the goal of the scoring model should be - at the very least - to never discourage people to play games. We absolutely can have negative points here, but doing something Elo-style is misguided and will only serve to have people game the system, which is a problem with larger player bases but unsolvable with small ones. You have to design a system to the reality of the variables you are playing with.

You don't even lose points for losing a game nor do I see how this could possibly correlate to "creating more larger games", take it from someone with years of experience.

It's very simple: it's more time-efficient in the current scoring to play games with more players, as you gain points depending on how many players you beat. When people were playing "ranked" shortly after it came out it did indeed cause many more larger games as getting points via 1v1s is simply not an efficient use of time (which had been a complaint about other ranking systems favoring quick, smaller games).

I know this will bug you, but your experiences are likely not relevant to the current situation. Like I said, experiences with large player bases simply doesn't matter to the case at hand here. What other games have you played with working ranking systems that have ~10-50 active players and no matchmaking? That experience might actually be relevant and I'd be curious if you can point me at such a situation. For my part I've had experience on all parts of these systems... the math, implementations, management and many systems as a player, etc. You're just going to have to trust me.

I guess if we're just going to keep the same old crap reset it at the very least...

I've given you the chance to come up with a scoring setup for a game that includes negative points or whatever else you want - the only constraint is that games are scored in isolation. Beyond that go nuts: since it is fundamentally a points metagame at this point, might as well change it up and let people rediscover how to game the system every once in a while, right?

[Asmodian]

I don't think people are arguing that an elo system would accurately reflect skill (at least I hope not). I think people just want a system that is semi-fun even if it doesn't mean much.

Here i'll create a point system that is super simple that people would like a lot more.

Details
You can't drop below 0 points

If you wanted to you could keep track of a players points for each games type (with the same guideline in place that you can't go below 0 points in a certain game type you have no points in). I know this would have a very minimal, almost not existent impact on a persons overall rank, but it's always fun to be able to break down your stats/points a little more to see them per game type)

quarterly or biannual full resets. I would go with biannual since there isn't a lot of games being played.

If someone abuses the scoring system oh well. That would be on the community not you.

FFA - Minimum of 4 players for scoring to count

1st place - 1 point for every opponent you beat - if there are 3 opponents you get 3 point, 4 opponents 4 points [If you have an ally or not it doesn't affect the rate you gain points at]

2nd place - 1/2 a point for every opponent you place higher than.

3rd - everything but last place - you lose 1 point

last place - you lose 2 points

2-team - Minimum of 4 players for scoring to count

1st place - 2 points

last place - you lose 2 points

tie - 0 points for both teams.


Icons

Every 5 points is an upgrade in an Icon

dagger: 0-4
2 daggers: 5-9
3 daggers: 10-14

You get the point...

For the celest ranks it's based on your total relative points to everyone else like it use to be.

10 celest icons:

10-7: Half moon
6-4: Full moon
3: Eclipse
2: Sun
1: Comet

Disconnects/Drops

-2 points for a dc/drop

10 game safety pad (your first 10 disconnects you don't lose points)


Anyways of course this isn't a good algorithm for scoring, but what it does do is it significantly rewards people for winning and losing. That in itself will create healthy competition that will bring more activity whether it is just # of games played on the server or an actual increase in players.

[punkUser]

I don't think people are arguing that an elo system would accurately reflect skill (at least I hope not). I think people just want a system that is semi-fun even if it doesn't mean much.

Yup, and I completely agree. I'm just saying it's misguided to think that something like Elo will increase the fun over something like what you posted for instance... it's actually the other way around.

I'll take a look at implementing your scoring below when I get the chance - looks pretty straightforward although you're going to have to clarify the scoring of ties in both FFA and 2-team - both whether you count as "beating" people you tied with and whether or not tied places count as a single slot or not (i.e. if two people tie for 2nd, do they both get 2nd place scoring and no one gets 3rd, or does the next highest team still get 3rd place scoring). Also min 4 players in FFA could still mean only 3 teams.

If you wanted to you could keep track of a players points for each games type

It's going to intrinsically do that as it does today in stats. Basically it works like it shows in the FFA league games for instance - it will assign a score to each player for a given game based on other factors in that game. Then later you can aggregate whatever subset of games you want (i.e. based on type or just overall or whatever) and sort to get ranks.

The "can't go below 0" thing is somewhat problematic though as that is persistent data and it actually makes it not a zero sum game. i.e. someone can continue to play against a "0" and expand the total number of points in the system. We can definitely not display someone as having less than zero points, but internally you can be negative... if you couldn't that would also represent scoring that isn't isolated to a single game.

To be fair this scoring isn't zero sum to start with so there will still be a general effect of rising in ranks as you play more games - which is what LK says he doesn't like - but I'm not really too concerned with that.

quarterly or biannual full resets. I would go with biannual since there isn't a lot of games being played.

Stats resets are trivial... them and changing scoring we can do whenever. It's not really so much of a "reset" as I just create a new scoring "season" and switch the active one to that.

Disconnects/Drops

In the current setup it's not really possible to robustly detect a host d/c drop... so while we can penalize players the host can always drop (well, pull the plug... if they just exit ingame it will still count) and the entire game won't be counted. Not ideal, but there's no good solution to that that isn't fairly complicated.

[Lord---Scary Owl]

kilg if FFA game were worth a lot. Then everyone would camp like malsleizure and they would be no fun. Everyone would gain crap loads of points off of being terrible and getting more time or sitting in a corner until 1 min left then using 100% vs 20%. 1v1's show that you can outplay people, and they show true skill. If you can beat a prestigious player and get less points than losing a lmoth to 3 campers after you've killed 2 players, thats a messed up system.

[Dantski]

kilg if FFA game were worth a lot. Then everyone would camp like malsleizure and they would be no fun. Everyone would gain crap loads of points off of being terrible and getting more time or sitting in a corner until 1 min left then using 100% vs 20%. 1v1's show that you can outplay people, and they show true skill. If you can beat a prestigious player and get less points than losing a lmoth to 3 campers after you've killed 2 players, thats a messed up system.

If you're killing 2 players in a LMOTH game you're not playing to win. LMOTH is all about trying to engineer a situation where you can secure the hill for 5 seconds at the end of the allotted time or in sudden death. If you just want to fight people play BC FFA or 1v1's.

[Lord---Scary Owl]

It's more like 3, FFA is real boring and shows no BC skill.
I like Body Count and ruining dbnova's games.
(And I still win the lmoth's or get 3rd in koths while doing so)

No one good is around to 1v1.
Rawr is sometimes there, asmo got on once but he is too afraid.

Hop and lets play some Dantski, winter myth is no fun.

[Lizard King]

Lso, winning an ffa has always been worth more. It's harder to repeatedly win ffa games by far than it is to repeatedly crush some newb in 1v1.

[Lord---Scary Owl]

Well it all boils down to...who are you playing with?

Beating someone superior to you over and over in 1v1 would be worth more than playing and winning FFA in vs some noobs.

BUT

On the other hand crushing a noob over and over would be worth less than winning an FFA vs FFA League Champions all in 1 game. I do agree with you there.

The main point is that the little icons don't mean anything. Like noobs wearing nova and Chohan wearing double axes.
There needs to be a system that would incorporate game wins, dmg, kills, and it would actually be cool to incorporate a 1-5 ranking system into the players account.

Like when you click view stats you can see name , 1-5 rank, quote, location, points, etc.


And off topic back to Milk Man's point having an Order tag would be cool.