Waiting hour for duel partner who will:
match your ELO
match your ping

QL needs more players first, before spending resources on coding the matchmaking.

QL needs decent matchmaking for more players to come. People tend to ragequit and uninstall when they try to pick up a new game and get raped into oblivion.

It's a bit of a vicious circle.

>>> and get raped into oblivion

But that's not really what happens. New players get dropped into tier 3.

And that's the average of the player base. (They've done their gauss curve thing, they might not be the smartest devs (according to esr), but they're not that stupid.)

If you keep winning you move up, else you move down. How else could they possibly do it?

(It's not perfect and tier 4 should be split up into more tiers imo, but it's not that bad. Newbies do not start in tier 4. Have you seen games played on tier 1? I saw a stream the other day, it's quite revealing.)

If people get turned off it's
A) because of too many idiots spouting a non stop stream of insults and expletives. (A lot of that comes from being obsessed by ratings and stats btw.)
B) to a lesser degree tier slummers and cheaters

For me the bottom line is that this type of arena shooter is just not liked as much anymore as the myriad of other games out there today. It's not 1999 anymore and there's a lot more competition.

That's the problem, sure, tier 3 might be the bottom-of-the-barrel noob but then what's the point of tier 1 and tier 2? You need to drool on your mouse and kill yourself repeatedly to reach it? Tier 4 is also a gigantic tier which has such an immense disparity of skill. I mean, you got rapha at tier 4... and me at tier 4, and we're not even fucking close; we're galaxies apart for skill. Yet, sponge still defends his retarded tier system like it's functional.

People get turned off because the game is difficult. If you want to believe it's because of people saying bad words and cheaters, you're fooling yourself. Hell, these new players wouldn't even have the knowledge and experience to know what a cheater is. For them, a great player would be considered a cheater, and we've all seen that before. Not to mention, even if the user is put into tier 1 or tier 2, there are so little options, the user is pretty much thrown into the wolves if they want to play. They promptly get raped and the uninstall begins.

QL needs matchmaking AND it needs a population. Of course, with the introduction to steam and a (potential) spike in users, this would've been the SMART course of action but instead -- to no surprise to many since sponge is a MORON -- they're going a different route.

Lets take this into an extreme example

Lets say there are 20 QL players total.
And four tiers.

10 of the players can't move and shoot at the same time.
5 can but are not very good at it.
You and 4 others can do both exceptionally well.
And then there's Rapha.

What do you do? Put Rapha in tier 4 on his own with no one to play? or scrap the 4th tier? Some other suggestion?

Honestly I'm sure there are people that understand matchmaking a hell of a lot better than me, and could even take a stab at writing the skill assessment and player matching, but I see a lot of complaints about it, but no one offering solutions that deal with the basic problem of numbers of players at specific skill levels, in very low numbers, or difficult locations.

What I would do is scrap the shitty tier system in general. It clearly needs a different set of tiers. Obviously, your example is extreme. Look at qlranks for a tier system that actually has some sense to it based on a statistic -- elo. Rapha would be GM, the other 4 that are exceptional can be M or maybe high level G (which would still play with GM's and high level M's), the 5 would probably be below that and the 10 would be at the bottom.

The tier system just doesn't have any relation to a decent quantifiable statistical property.

Maybe I'm seeing this wrong, but a rating system is not the same as a matchmaking system.

And the average newbie will not start with duel.

A matchmaking system obviously has to use a rating system of some sort, to match a player to another player, you realize this, right? So whether they are different, and they are, they'd still work hand-in-hand. You'd need some legitimate ranking system to have a proper functioning matchmaking protocol.

Well yes, that was the point I was hinting at.
Skillrating or Elo, you still need a matchmaking system.
(Also this matchmaking system has to work for all game modes, not just duel, just mentioning to make sure.)

You make it seem as if replacing skillrating with Elo solves everything, but it doesn't, you're still stuck with the same matchmaking system they have now, or something extremely similar.

(Improving the tier system: yes let's ad an extra tier: but will that solve everything? I doubt it.)

But QL actually has a very similar system in the background that's used to split the tiers.
QL duel is actually Elo based.
They just chose to divide the numbers down into a more generic 'skillrating' and keep that hidden, then split.

tier 1 safety net then the remainder split -
25% lowest
50% mid
25% top players.

if you used QL ranks alone, at what point would you prevent a player from not ever being matched up with another?

20% gaps so 5 tiers?
10% gaps?

But what happens when the bulk of the players in your gaps are continents apart?

Only the QL devs have the real numbers at hand, they dropped from 5 tiers to 4 for good reason once already. If they thought it would work I'm sure they'd hike it back to 5, which would be good really as new players could be inserted right in the middle tbh, but I'd still be very concerned about the number of games available.
QL's highest popularity is still in the EU, and the 5th tier was dropped simply as even in the EU tier 5 games JUST DIDN'T HAPPEN, and that was 3/4 years ago when everyone seems to think the player base was larger.

Maybe a small tier increase will happen for steam as the global numbers should hopefully increase.

But I just don't ever see some miracle algorithm ever appearing that will solve things the way people seem to think.

QL doesn't use ELO. I can assure you. The tier split is as ridiculous as I pointed out. I'm tier 4. Rapha is tier 4. We're not even close to being in the same galaxy of skill in duel, yet there we are...

If a matchmaking system were based around the current ranking system of tiers I'd be playing against Rapha. Doesn't make sense, does it?

The thing I find funny about people is that they give developers the benefit of the doubt. How often do we have to hear "guys, they're the ones developing the game, they know what they're doing." It's almost like a parrot/sheep mentality. Here's a clue... the devs don't know SHIT. Quake was luck in the first place that it became such a godlike FPS/internet sensation. The developers don't have a god damn fucking clue as to what they're designing in the realm of any competitive aspect. Seriously. I'm not joking at all. Sponge is a prime example of a clueless moron who has no skill, experience or knowledge in Quake AT ALL. You don't need to be Rapha to understand the nuances of Quake, but you got to have a fucking clue, and he doesn't.

You're getting way too specific on how I'd run a ranking system. I haven't sat at home and fleshed out my DREAM SYSTEM on a napkin. However, simply using QL Ranks and an ELO system would've drastically improved all mods. Percentages and gaps and all this stastical shit... I don't know, but I can guarantee that because this community is full of awesome, smart, intelligent players, they could easily design a framework for some sane system.

To summarize, the QL devs don't know DICK. FOR REAL. They're idiots. Clearly. To even defend them as such is fucking mindblowing.

EDIT: And let me reiterate, I'm not saying a matchmaking system would SAVE Quake as one that works needs PEOPLE, but when you're finally bringing the game to Steam, and would GET a surge of players, this would've been the best, smartest and most opportune time to have one ready for the masses of new players.

QL doesn't use ELO. I can assure you.

...and how can you assure it?
Duel skillrating is Elo based. Not just win/loss though, it takes frags into account. Moreover, tiers aren't used in duel. Well technically, duel servers still have upper skillrating limits which are related to tiers but it doesn't affect the "Skill matched/ Your skill higher/ More difficult/ etc." thingy. Which can lead to funny situation where you can sometimes see a "Skill matched" server which you can't join cause your skillrating is too high for it.
But yeah, the little checkmarks are up/down arrows to show skill difference are purely based on Elo difference in duel, not tiers. Which means you can be tier 4 in duel and still see red duel servers. You can test it.

Because sponge/Sync have admitted that their system doesn't use ELO.

EDIT: You realize you can have other systems in place to define tiers, right? They don't use ELO.

Because sponge/Sync have admitted that their system doesn't use ELO.

source?

IRC.

Here's one IRC source for you :

<@sponge> thw duel is elo based on expected win gap

Where's yours?

<@sponge> duel isn't based on elo

There.

<toxjq> hi guys how to use EL GE wepon shooty laghtning q:^^^^^^^^^^^^^D

Funny I can't find it in my logs. I guess I must have not been in this channel at the time!

***EDITed the rest out cause I'm not gonna step into the grey zone longer for you.

Quotes please, as they've always previously said duel is the exception to the ratings used in other modes, and uses an Elo based system.

http://www.esreality.com/post/2281167/qlranks...mparision/

What now?

Were you just proving that QL doesn't use ELO?

Do I need to link you each separate comment that explains how QL does use elo and why you may be under impression that it doesn't, or are you old and capable enough to skim through a thread with 60 replies?

I am too old and don't care enough to skim through a thread with 60 replies.

How about just http://www.esreality.com/post/2281167/qlranks...pid2283409 then.

Lying to shut you all up.

That's all it takes to end a discussion?

So we can now roll back to this comment http://www.esreality.com/post/2642244/re-ql-m...pid2641652

And just claim you've been lied to.
Done.

It's mostly apathy at this point. The changes will go through, or maybe some slight changes, but Quake, as we know it, will finally die. The people defending sponge/Sync, or the changes, mean very little.

It's moot. Like megaman says, even if they DID use ELO, it's clearly awful, based on your own thread.

But how about this for a finisher: you win. I blew my load on how dumb sponge has been, I really don't care any more to defend my position, nor comb through logs to prove it. I'm spent, I need a cigarette.

dogs bark, but the caravans move on.

quake -it never realy dies.

What about the other mods? Duel is at best 10% of QL.

Also whether they use elo or not, their system is still crap in duel. You have 1300 qlranks elo players being paired with evil (2400+ elo).

paired how exactly?

As for the other mods, they use a... peculiar system. In practice it does pretty well as long as you're not in the very low or very high ends of the skill curve but let's just say that it's not based on a superb mathematical model like TrueSkill. Players over 100 in skillrating typically aren't gonna have a very accurate unless they played enough AND that they mostly play against other players around their skill level. So yeah, it's good enough for the majority but kinda sucks for the best players.

The system is simple. At the end of the match, a value is computed for each player : their score divided by the time they played in this match (or the other way around idk exactly). Players which haven't played for long enough are ignored. The rest are sorted in a single list according to their score/time value. The top 1/3 gain N points on their skillrating and the bottom 1/3 lose N points. I don't remember what N is... it might even depend on other stuff like how many matches were played by the player so far maybe or perhaps it's different among the same group according to the internal position there...

OT but TrueSkill is the same as Elo - in case anybody was wondering.

"It is an extension of the Glicko rating system", which itself "was invented as an improvement of the Elo rating system".

http://en.wikipedia.org/wiki/TrueSkill
http://en.wikipedia.org/wiki/Glicko_rating_system

Yeah, it's like saying racing cars are the same as steam engines.

Don't get cute.

It's like saying a Volkswagen is not good enough, you need a BMW
when you need to move some boxes.

They are completely different algorithms (Elo and Glicko). Both invented to solve the same problem originally, but different in approach.

"Moreover, tiers aren't used in duel"

Then how can you explain the following?

I'm 1500-1600ish in duel in general. I made a new account, where i absolutely don't care about stats, just pure practice. I haven't played much games yet, but if I do, I play 1500-1800 guys whenever I have the chance, so my elo is 1400 now, with tier 5 (gold). One of my friends (1700ish guy + really good aim) made a server and it was GREY for me, more than once.

I posted this back in 2012, has some explanation

http://www.esreality.com/post/2281167/qlranks...mparision/

You are tier 4 (gold is just a graphic thing).
Your friend is tier ?.
You friend's server is grey for you.

Nothing can be inferred from that except that your skill estimation must have been higher than your friend's­. I fail to see where it challenges the claim that tiers aren't used for duels.

"your skill estimation must have been higher than your friend's"

Based on what?

Based on the fact that the game tells you the server your friend is alone in is "Your skill higher".

Haha. So you have no clue.

tier 1 safety net then the remainder split -
25% lowest
50% mid
25% top players.

wait what? does this mean that the worst 25% of players are in tier 2, the best 25% are in tier 4 and the remaining 50% sit in Tier 3, effectually making it twice as big as the other two? And Tier 1 is an essentially meaningless placebo?
Or is it
0-25% Tier 1
25-50% Tier 2
50-75% Tier 3
75-100% Tier 4?
But I'm pretty certain it can't be that, cause I'm sitting in Tier 3 atm with ~1100 ELO and I have to be one of the worst active players all around. The notion that there have to be at least as many worse players as there are better players than me is quite frankly ridiculous, unless id don't remove inactive players (=not played in X weeks or so) from the ranking. But surely they do that, right?

As I described it is how the developers detailed it to me.

In fact this post was made about it at the time, ash just didn't mention tier1.
http://www.quakelive.com/forum/showthread.php...hancements

That's the whole problem with the current system... it's retarded.

What's your reasoning for thinking that?

by far the most complaints we see are

tier3 players that get good enough to make it into tier 4 who want an easy ride again.

with a fairly distant follow up of
tier4 players who want low skilled players kept out of their games.

I don't think i've seen a single comment from a tier2 player upset they've moved to tier3, or tier1 upset about moving to 2.

when i was playing i would wary between around 20 to 40. once i won 2 games and was placed in the same skill bracket as amai, who im pretty sure won zotacs.

Because tier 1 and tier 2 don't exist. That's the whole point. Just like in MMORPG's, the biggest population of players is level 1-10. Doesn't mean those players are actively playing. The ACTIVE population is probably primarily composed of tier 3 and tier 4.

And earlier in this thread an 1100 elo rated player is tier 3, but then I'm at 1500 elo rated player and I'm tier 4. Then there is Rapha at 2500. You see the difference and the problem with their "elo-based" system?

You're complaining about an issue inherent to Elo itself....

Typical Elo distribution graph :
http://lolmatches.com/charts

I'm replying to your "you don't hear complaints from tier 2 and tier 1 players", that's because they don't exist, whether it is an inherent issue with ELO or otherwise. Hence the bigger issue of having 4 tiers. If the first TWO tiers (50 fucking percent) of your whole ranking system is moot, pointless and non-existent, you've fucked up. The biggest problem is that the 4th tier encompasses too many people.

I posted the distribution graph to show it doesn't matter that Rapha has 2400 elo there is going to be too low a percentage of people near that to make a tier for them, Elo gives very skewed ratings, you gain much more for being near the top.
--------------

What makes you say the first two tiers don't exist, when you've just been told that tier1 is the lowest skilled 10% of the userbase and tier2 25% of what remains?

It's not taking 32.5% of the lowest possible skill ratings and setting the tiers to that even if only 3 total players fall into skill ratings that low.

tiers 1-2 hold fully 32.5% of the total player base, they very much do exist.

I'm not sure you read the ACTIVE part. Just like MMORPG's, the biggest population doesn't necessarily equate to the most active. Thereby, a tier 1 and tier 2 player, who actually plays constantly, would reach tier 3 without much issue.

Therefore, a tier 1/tier 2 doesn't really "exist", if only to be part of statistics. Of that 30ish %, I wonder how many have more than X number of games. I'm sure very few.

There are more tiers in duel, they are just hidden. Even though I am tier 3 I get "red servers", which means, that they're 2+ tiers higher. And it was same when I was in T4.

T1 has around 10% of the active playerbase : the worst of the worst. They are really extremely bad. T2, T3 and T4 respectively have indeed 25%, 50% and 25% of the remaining active players. So that would be :

0-10% Tier 1
10-32.5%% Tier 2
32.5%-77.5% Tier 3
77.5-100% Tier 4

Please note that T4 players are much more active than T2 players so there's more activity in T4 than T2. The proportions here describe the amount of players (or accounts), not the proportion of activity in each tier compared to the total activity.

Thanks. You mentioned in another post that the tiers aren't used for the server list in duel and that the order is actually based on an ELO difference. Does this mean the tier I'm in is meaningless when I'm trying to find an evenly matched duel and that the info provided by the list ("very difficult" "your skill higher" and so on) is essentially a less precise version of the QLRanks script?

I guess it depends on what you mean by meaningless. It's mostly meaningless. Both your tier and your "skill difference indicator" are based on your skill estimation but they don't affect each other.

What do you mean about the QLRanks script though? Is there a script that does a similar job as those skill difference indicators? In any case, here are two bits of information :

- I'm not sure how the "skill difference" is determined. I think it's based on the best player in the server rather than the average of all of them.
- I much prefer QLRank's skill estimation. It's far less volatile and it's purely win/loss based instead of trying to do some voodoo with expected scores.

certainly skillrating shown with /serverinfo on the duel servers shows the SR of the top player, unlike other game modes where it's an average.

The tier I'm in is meaningless in the sense that it doesn't affect the difference indicator at all.

The qlranks script shows me the ELOs of a server in the server browser. Since the skill indicator is also (partly) ELO-based it's kinda the same principle.

I agree about the superiority of qlranks. We all have seen the apparently "skill matched" servers in the list with players of >400 ELO difference to our own skill.

imho they're counting inactive players too (as they do when they say 1% of players use non-default picmip) which creates an inaccuracy here

1100 elo and you think you're the worst?
Boy are you wrong about that :)

"<@sponge> tier 1 players dont rocket jump, or fire their guns while moving"

I've seen Tier 1 action and they are still figuring out how to use the mouse and keyboard at the same time. Not even kidding.

"tier 1 ACTION"

: D

Matchmaking:

To my understanding, there is not a -clear- matchmaking system currently in progress, as there are no visible tiers for the user to see. In general, there are also ~5 tiers in total, which makes very little room to move up or down the rankings without it being a significant jump. Users need to be able to see their own rankings and not just who is below or above them, as that holds very little value to the player. Everyone knows there are 'good servers' and 'bad servers' (more challenging/less challenging/same skill tier), but there needs to be specific number skills on individual users, and a much larger diversity in the number of tiers.

~Slasher

yes you put rapha in his own tier, you don't sacrifice the entire playerbase so a rapha level player can play.

That's exactly what they did. Sacrificed 99% of the noob playerbase so the 1% (elites) could play.

Tier 4 is a bit of a problem I agree, but not the rest. It's not as you described imo.

I lurked the official forums for a while. I only saw people complaining about "matchmaking" when a tier 3 players gets moved into tier 4 (and yes, get raped, at least at first). Sometimes there was a thread about it every day. I'm not exaggerating. Only tier 3 moving into tier 4. Never about newbies getting raped in their first games.

A dev once said they didn't have a big enough playerbase to split tier 4, but I have a hard time believing that. Also if tier 3 is the middle: there's 2 tiers lower but only one higher... So imo there's room for at least one more tier.

If you are strong enough to move to tier 4 after say one week, it means you are not really a newbie when it comes to arena shooters like Quake and should be able to get better soon enough not to get raped so much anymore. But it's still the biggest hurdle as imo tier 4 is too big. As you said: it pretty much includes everybody who more or less knows how to strafe jump. And that's a group too large to throw together (including casuals and pro's like rapha, cypher etc).

(Starting everybody in tier 1 is not a solution either. The average new player will rape players in tier one. Which shows why matchmaking is not that simple as it may seem.)

Why not have a player go through some ranking matches? Hell, even if it's against the computer, it'd properly be a lot more effective for placement than we think. If you're competent at Quake, you could hold your own against the hardest difficulty and could be slot in at a higher tier (assuming we're still on our shitty current QL tier system, let's say 3), and if you get stomped, you can start at 2.

Obviously, just a rough idea but what's the problem with having players go through a few "ranking" matches. SC2 does it.

Do you mean something like the compulsory crash course QL used to have?

Ehhh, maybe. Jumping skills are important but I guess I leaned more towards a quick first to 20 hardest difficulty 1 on 1 against the computer. But again, I'm just throwing out ideas. Maybe a combination of the jumping tutorial and a battle tutorial would be even more effective.

Shrug.

EDIT: Or even telling the user they're going to do 3 ranking matches in duel. Or 3 ranking matches in Clan Arena might be friendlier and at least allow players to potentially make some kills.

>>>Jumping skills are important but I guess I leaned more towards a quick first to 20 hardest difficulty 1 on 1 against the computer. "

Uhh that's what the crash course was, somthing tells me you weren't even around when it was mandatory

My Quake Live account is wormed. Go check the creation date.

clearly photoshop

Why not have a player go through some ranking matches?

They do. The first games a newbie plays are his ranking matches.

But the initial ranking is not really the problem.

As I originally said: the complaints are not from new players getting raped in their first matches but from tier 3 players who get moved to tier 4 at some point.

(At least on the official forums. You may argue that they do not reflect reality well enough, but then we move into the realm of speculation without facts.)

I don't think new players that get raped in their first match and quit are gonna go visit the official forums to tell the world about their humiliation heh.

I don't think many players get raped in their first map.

There's t5 (aka gold) after t4.

Not that this tier-bs would mean anything tho.

No there is no T5, the gold thing is just graphic.

Certainly different than plain t4 in terms of match ranking.

What matchmaking?

Need pro sub for that.

wat

)

I always get a game, and I never have a problem with my opponent. duno what's wrong

Region

de

and that's the reason why you don't have problems while other complain. :)

losing is part of the game, deal with it

I am playing in de servers. ;)

I want some.

It's a technical nightmare, would never ever work. Think Ql 2.0 is bad? You can't imagine what trying to implement matchmaking would do. Stop even mentioning it. If you want to know more, ask Sponge @ Quakenet for details.

Ideally it would be exactly like csgo's I guess

My analysis of it is the following :

1. Don't wait for too long. It's fine to be put in the pool of available players and wait a little to get the best possible match, but don't overdo it or I'll get bored and leave.

2. Don't send me too far away. I really don't want to play with more than 80 ping. The lower, the better. Different players have different tolerances to ping to complicate things.

3. Don't choose the wrong game mode. Well, this one's obvious. I wanna be able to choose which gametype I'm gonna be playing.

4. Don't match me with players out of my league. Not too good, not too bad. I want everyone to be pretty much equal. It also makes the process of team building much more difficult to screw up.

5. If possible, let me play with my friends in priority.

If I forgot any factor, please point it out!

These expectations translate into constraints for the developer's matchmaking system. You have to worry first about not leaving anyone out, like by giving rapha a duel opponent which will do when no good player's available, even if he's much worse skill-wise, Then you worry about selecting the best possible groups with the remaining players.

It's like some kind of constraint satisfaction problem in multiple sets in a non Boolean environment (OK, not OK and all that's in between). Sounds fun.

but anyone who would be in rapha's range, even those in the lower end of it, are already good at quake and have accepted the game as it is. the problem is more about beginners getting smashed by mid+ players, not good players losing to great players

you really dont see a 1900-2000 elo duel guy crying about losing to 2500 elo people

It was the simplest example I could think of to illustrate the concept of a constraint in this context. The problem doesn't exist only with skill and it is more complex when you go above 1vs1. I'm describing how the system should work, not how it can help newbies.

But, for the record, all matchmaking can do for newbies is to avoid matching them as much as possible with known skilled players. It cannot separate straight away bad beginners from skilled beginners. It will match these newbies among themselves and long term mid skilled players. Yep, just like how it works with QL right now in Tier 3. The difference is that a matchmaking system should be scalable and do better the more players there are... so it does better than the tier system if there are enough players to allow it. Oh, and also in QL players can stay in a T3 server even if the result of their last match there bumped them to T4. That's bad. Depending on how you do your matchmaking system you might just end up kicking everyone and putting them again in the available players pool to create a new group of players and you avoid the problem that way.

No-one has a solution, yet all demand it. I doubt Quakelive has a large enough user base to make duel match making viable.

I think most imagine it a thing that should go with promotion, QL 2.0 or steam release. Would have worked wonders with the release out of closed beta.

It can even be disabled if there are not enough players online.

Coming out of closed beta (Or was it open?) QL had a "find match" button. It was dropped pretty quickly - main reason being it wasn't scalable enough to be able to cope with the large influx of players, afaik.

I guess with low numbers they could re-enable it if they want, but sponge can see that it's not going to work out anyway, and re-engineering in a way that can cope with going to steam has clearly been estimated.

and re-engineering in a way that can cope with going to steam has clearly been estimated.

You mean, you think they're planning for matchmaking to be done for Steam? Why do you think that?

No, the scale of the work estimated, and then rejected as "we don't have resources for that" (Assumption based on the chat logs, since I haven't joined the focus chan in years now).

They have a working(ish) implementation (for duel). They will definitely have evaluated whether it is possible to roll it out and improve upon what they did way back then.

Gotta envy those small scale projects that do have the resources for that.

Such as?

bloodline champions

Maybe small, but judging by the "team" page they have far more programmers than the QL team. You gotta remember that QL is not a priority for id (shipping Doom is presumably #1. #2, and #3 in their todo list)

Firstly, I don't feel like sponge is accurately estimating the hardship of creating matchmaking. With the amount of work they did on other stuff / will do on other stuff, they could have had matchmaking ten times over. Priorities.

Secondly, making complete products and selling the right ideas to executives, I'd say, is a skill such a company should have. But who am I to talk, let them run Quake into the ground.

I think sponge is probably right that it is completely out of their budget to implement given the resources they have, they would need an explicit new hire / contractor.

I think there is some argument that perhaps it could be justified with the steam launch that it is worth trying, but given the complete failure last time under (hard to test) high loads, it would be a very hard sell to management that it is worth the money and could be a huge embarrassment if it failed under the loads again.

I agree with sponges assessment that matchmaking cannot help players who give up after 0.5 games given matchmaking cannot correctly match them without a playing history (although I think you could perhaps do more by deliberately giving them artificially lower ranking than he is assuming), and he is in possession of the stats of how many people do give up after that time.

I think you're forgetting how much in QL is already there.
They have an infrastructure that shows each different client "which server is equal skill" (which atm doesn't make any sense).
They also have a stats system which tracks all results for all players.
And they have that "play now" button already there.

There has to be a fair way to make those 3 things work together in a way that wouldn't march noobs to the gates of hell on their first go. And possibly it can be implemented in a matter of days rather than months seeing how all those things already work.
It may be a bit hard on their server though (depending on what solution you come up with) but it would be worth a lot to actually gain players, don't you think?

It may be a bit hard on their server though (depending on what solution you come up with)

That's what they tried before, and it brought the whole system down Iirc.

I don't imagine they want to bring back the "queue live" award for the steam launch...

Edit: not saying the find match button was the only reason QL couldn't cope with the numbers (the stats and ladders went offline while they made those more scalable), but the fact that it has never come back is telling IMO.

The way I see matchmaking, it's simple enough. Perhaps with shoddy coding, silly (it is) skillranking and millions of database queries for stats online, it can get troublesome... The only thing I can't admit I have a good grasp on is how much resources it takes to keep the whole thing online.

Still, I can't help but feel that they did matchmaking without elo or anything similar, also did shoddy rankings, and then put the whole thing down and blamed it on matchmaking being not viable. Doesn't sound like problems simple matchmaking causes.

well in general... if you assume you have the numbers (userID and elo for instance) then it's just finding integers close to it.
Most algorithms used to find integers close to eachother are in an order of magnitude of (N^2) where N is the number of records you have.
So if you have a list of 1.000 people trying to play a game at one point in time you'd have to make at least (1000^2) 1.000.000 calculations. Now given that most machines run around 2~4Ghz and you need multiple cpu instructions to actually access the numbers, and you probably want to check on multiple things as well while you're trying to match players and accessing memory takes a bit of time....

My estimate would be that you can do all of that in about half a second on any normal machine.
But ofcourse it's given that you actually have 1000 players wanting to play at the same time...

I think it's more than userID and elo, there's also how sure the system is of your skill. There's recalculations of those. And there's ineficiences with databases or some shit, dunno.

I still can't help but feel like it isn't huge.


Aaaand it would be implemented by a fan for minimum wage or something if they let 'em, as long as bringing people up to speed to QL's other code wasn't required / a problem.

Well in general you try to not look at how sure the system is of your skill while matching. You do that when you rank them (or attribute elo to them).

And you don't have to know anything about QL's code...
You can just write a completely external class that does ranking which just assumes that you "get a userID and elo" from somewhere. Then list those in some smart ways and output an array of lists each of which contains a match.

When you want to "add it into QL" then you'd need to know which function you need to "give your lists" to though and in what format etc.
But I guess that if you do have access to all of their website code... it's not hard to actually write a webservice which recieves a list of players and a gametype and gets all of them to the same server.

Things like this are very modular in nature. You just need to be fed the right information and then output it in a correct way (which the recieving party understands).

That said there are several things which you do need to take into account which I'm not sure of how they work in detail (I'm not a web-developer). Which deal with how you get information from one place and use it in the other. In general to access another machine takes quite a long time (due to ping and security issues etc) so you don't want to continuously set up a new connection but rather just send stuff across in bulk. There are some problems which could occur there which you'd need to know about (I guess most web-developers do know that stuff....)

When I said how sure they're of a player's skill, I meant a number of how sure they are for each player. Kind of like in MS' TrueSkill, you can read it up if you want.

Either way, we're not actually disagreeing on anything here I think. I just don't understand for what reason people think matchmaking is actually problematic technically. I could miss server / resource problems, but I can't figure out actual technical hurdles.

Well... I do know most technical aspects, and I also can't figure out what's so difficult about it.
Ofcourse to build a complete system from scratch which users can use... that would take a lot of work.
For instance you could spend an entire day just making the button look good. And you'd need quite a while to actually implement something which forces players to some server. (that is actually quite technologically advanced how QL did that from a webbrowser)

But those things already are in QL (or well they were at some point in time, so the functionality is there).


To comment on the uncertainty... Well that's TrueSkill ye... but you can't really copy that into a commercial product without running into trouble ;)
Most systems just work with a single average value which they compare to the average of others... But sure you can add more stuff to it... but in my eyes it would make it needlessly complex. A "more default behaviour" would be fine for the majority of the generated matches.
The more stuff you add to it the more complex it gets... so just start fairly simple and see where it goes would be my plan.

I feel like some variable on how sure about your skill the game is isn't copying TrueSkill :) But perhaps it could work without it, I don't know.

Well let's say that there are many ways in which you can make a somewhat different matchmaking methods... But the big step to take is to actually have a somewhat decent/working matchmaking method.

Definitely.

Well... I do know most technical aspects, and I also can't figure out what's so difficult about it.

Excuse me...

Most algorithms used to find integers close to eachother are in an order of magnitude of (N^2) where N is the number of records you have. So if you have a list of 1.000 people trying to play a game at one point in time you'd have to make at least (1000^2) 1.000.000 calculations.

Have you heard of sorting and binary search? Self-balancing binary trees? k-d trees?

How hard can it be to create a webservice which takes as input "userID, gametype he wants to play and his elo/rank" then stores this in a hashmap or multidimensional array.
And each minute, or just whenever there are enough people of similar elo in it, it gives a list of matches linking the "closest people in elo wanting the same gametype".

Here's the problem, these data structures don't have an efficient operation to find the nearest neighbour of an element. Can you explain how will you do it (no brute force)?

Ofcourse I've heard of binairy trees.
You can search in them in log n and most other operations are n log n. But they are not very well suited for this problem as re-ordering them can, possibly, take quite a while. And in general sometimes you find a match and sometimes you don't so you want to remove parts of the tree. Also you don't know which side of a tree/branch the closest match actually is. Unless you want to make a pre-defined tree and put logic in that... but that would be really hard to do.

Anyways the easiest solution I can come up with is to simply use a hashtable per gametype.
So each time a new player hits the "play now" button you simply add him to the hastable with an index of his elo (aka addToHash(elo, playerID) where addTohash takes (key, value) ).
Then in order to find the closest you can just search it linearly and you can even put a max-range on it.

Worst case it takes N^3, but N^2 is more likely, where N is the size of the hashmap. Considering that atm there, to my estimation, are not more than 200 matches starting each minute you can easily put it to 1000 for all gametypes and run through it once every minute.

p.s. a hashmap already deals with issues like adding multiple people to the same index in a way that puts them next to eachother.

There are a lot of possible implementations btw, I can also imagine a really valid use for multiple singledimension linked lists.
Where you let each 'list' you have be an area in the world and you bind objects describing the elo and "gamestate" of certain players to multiple list.
That way you can easily check for games "near you" while not being restricted to just a single area.
Also it's extremely scalable because of the linking.

The adding of new entities into the list will be somewhat longer though (as you have to search where to add) but then "finding the closest in elo" would be very easy and you can actually directly check upon trying to add into the list.
If you make the class function which gets/sets the gamestate as a "syncrhonised" function you even lose all of the problems of being added onto multiple lists.
Having to wrap all the clients into a list would take up a bit more memory though.

Theres quite a lot of funny solutions to this problem :D

I thought they took it down because it didn't work as intended...

Anyway I would find it very strange that their machines would be unable to process 1 request per user per 10 minutes. While they can store detailed information on each user per 10 minutes.

How hard can it be to create a webservice which takes as input "userID, gametype he wants to play and his elo/rank" then stores this in a hashmap or multidimensional array.
And each minute, or just whenever there are enough people of similar elo in it, it gives a list of matches linking the "closest people in elo wanting the same gametype".
You'd just need to make it such that upon completion of the list it calls the already existing infrastructure of creating a server and guiding the players there for each generated match.

Sure it's not something I can make overnight... but it won't take me nearly as much time as making a new (decent) map or making several small mods or fucking up the movement system.

(btw if they want to have elo they can probably make a deal to copy it from the qlranks server and store it on their own machines)

The problem is quake is such a diverse game and even qlranks elo matched games can give scores of 20-0.

I was under the impression that the duel picker technically worked, but had issues because of limited players per region, but and the biggest problem was FAR too many players expect it to give them either a win with little effort, or games with only 1 or 2 frag margins all the time. And so was removed because the end result was negative PR.

FAR too many players expect it to give them either a win with little effort, or games with only 1 or 2 frag margins all the time

It happens in basically all the other games out there. Many games with matchmaking have much more issues with it than QL, I'd say. And nearly universally, matchmaking works out for the better as they have something to play for and improve in, and it makes starting another game just so so attractive, even if they flame it at first.

Look at Dota 2, or LoL, or any other game with matchmaking. Most of the public opinion about matchmaking is negative. Doesn't mean the net effect of it is negative.

Never played them but looking at the type of game they are surely you get a better experience in them as you can spend time building up things a while before you get stomped.

Whereas in an FPS you can end up literally feeling like you're not even playing just constantly dead.

That's true, but it's always true for Quake, and it's somewhat true in CoD and Halo. It's onyl easier there because you can get lucky kills.

But it's true with the custom games system too. Getting matchmaking for Quake wouldn't hurt it or defeat matchmaking's purpose. If the matchmaking was for some of the less hardcore mods, like CTF, you'd have the same kind of thing like in Halo.

Well at our level that's somewhat true. I'd be more inclined to say 10 - 2, as that's about the score I'd expect when I play an equal skilled played.

But I am inclined to think more about people that genuinly suck. Matchmaking for somewhat decent players is usefull but they can in general already understand why can go heavily into one side so they accept it and move to a different match / try to beat him the next time around.
But when both sides are kinda bad, this isn't nearly as obvious. Both players will mess up a lot of chances and not know when they can push. I've actually seen several games end in scores like 6 - 5 between people who could barely walk around the map in a straight line. And they had a great time doing so.
(it's not one of my hobbies to watch those games though but it has happened)

Interesting. I was expecting you to give me a steam exclusive title (with matchmaking through steam). I'll admit I don't know much about this game though.

(if the whole team we can see on http://www.stunlockstudios.com/the-studio/ worked on the game though then it's still 12x QL's current team size)

I don't think that there's "matchmaking through steam". Anyway, workforce isn't the point. Priorities, selling the right stuff to higher-ups and crazy estimation is. Whichever way, they'd have matchmaking ten times over.

steam actually does have something which games can somehow use which does cheat detection and matchmaking... I'm not sure how it works in any detail.
It's called "steamworks" http://www.steampowered.com/steamworks/gameservices.php.

People have wanted it for what, at least 5 years?

Now, it's going to steam. Steam users will demand it even harder than we do. Because their games have it.

So, they demand it and get it or demand it and don't get it and it's GG for QL

Matchmaking would work based on AIM and damage DURING THE GAME.

Let's say I join a tier 2 game and aim hitting and raping everyone, the server itself should zero in on my ass and go 'wtf nigga? why you up in here? shit son, fucking slumming ass'

And then I'd be slapped with a handicap 50 for the remainder of the game.

If the newb team is getting close and destroying my team after the handicap 50 is placed on me the serv then goes

'you ain't shit now eh nigga? ok then' and then puts hanidcap 70 on my ass. And then the game ends 10-9 for my team.

Newbs have fun, pros get some challenge. ALL IS HAPPY.

Except it has nothing to do with matchmaking.

it would make it so that everyone can play with everyone and not rage.

I'm 99% sure a ton of players would rage because they are punished for playing well heh. Also I've been playing on T1/T2 servers with a handicap around 20 and let me tell you and it turns out that if you wanna win, you don't end up improving your game in a way that would help you against good players. You only end up playing like a fag to avoid all kinds of damage because it takes so fucking long killing someone even with 60% lg that you're gonna be dead before you can do it if you're not super cautious.

As an idea though, some kind of auto handicap based on relative skill does perhaps have some merit. In the absence of better matchmaking than the tiers system, you could perhaps balance things by using ELO or whatever measure of skill you have handy to adjust it.

Sounds interesting (in general, not necessarily as a matchmaking alternative maybe) but would it be doable?

If the server reacts fast your handicap changes after every fight (and the noobs be like, WhaaatDAfuck is happening).
React too slow and game is already almost over.

Also you would be punished if you make a couple of good shots. You want that?

That's how I play in na ca, if i'm just dominating scrubs then I handicap myself. Makes it hard

I like the idea of dmg dealt and received during game, should be one of the parts of matchmaking. Perhaps last man standing in CA would also be used?

Aim could be exploited, by random hitscan spraying everywhere. Especially if your not concerned about your stats, but it would mean that they could remain unhandicapped.

That is not really rocket science. I want to specify time, location, mappool and skill range expressed in elo by qlranks for players with more than 50 matches.

Matchmaking is irrelevant. The time to explore that idea was when the game was first available to play or when the game released from beta. I give an exhausted sigh to anyone who believes matchmaking, Steam, ladders, anti-cheat, rule sets, etc. will make any difference.

I think Quake is on its last legs no matter what, I agree, but it'd have been nicer for Quake to die being the best FPS game ever, not the pathetic farce it's being turned into.

Matchmaking + Steam could've MAYBE kept it alive long enough for something better to come around (Reborn, Reflex, etc).

Newbs don't play duel. They will likely play ctf or ffa.

How many times in FFA do you see a newb in a server with 2000+ elo people? It happens way too often. And when I say newb, I'm talking about the guy that runs in a corner, stands there facing outward and tries shooting people with mg. The guy that can barely move. The guy that hasn't even figured out how to strafe. Those guys should never end up in a ffa server with guys that have been playing 10+ years. I barely play ffa and it seems I always see one or two of them per session. That's what needs to be fixed.

When I started playing Quake 3 in the test days, it seems like everyone were newbs. We sort of grew up together. I remember playing CTF in a tourney where demos were being recorded, and the demo showed me rjing for a boost of speed during a flag run and the people in the quake forum I belonged to that watched the demo of me had never seen such a thing (and I thought I invented it at the time lol).

Also back in the early days, in the FFA games people would always complain that certain people never died when they hit them with 3 rockets and blah blah blah. They hadn't realized yet that it was because of armor. Also there were ffa servers with 30+ people in them, most of them with 40FPS and 200+ ping. That was the setting for most veterans who have been playing this game for so long. It's how we learned. People now have a much different experience when starting out.

FFA elo doesn't mean anything under 2600ish.

The new 3-tier system with QL 2.0:

Tier 1, "Wait, how do I shoot?"
-Quakelive standard

Tier 2, "Wish this game stopped jumping for me"
-Quakelive "Classic"

Tier 3, "The rest"
-Quake 3 Cpma

Note that the tier 1 experience will include a warm welcome hug by Daddy_ .

of course there needs to be some waiting time, but the mathematics make these wait times much much smaller than the server-based approach of today

today if I want to play CTF or Duel or TDM, I have to basically join an empty server at 1 place for 1 of these modes and hope someone comes

the biggest win of matchmaking has nothing to do with skillratings, but rather being able to queue up for multiple modes/servers at once



anyways, the wait times don't need to be instant because all of the matchmaking should happen in the background while you are playing casual modes that you can abandon mid-match (eg. FFA and maybe clan arena)


something like:

1) let people choose what modes/maps/servers they want to play on (by default all maps are selected. by default all servers <75 ms latency are selected)

2) find matches. require much closer skill levels the smaller the game is (eg. a 4v4 with players from 1400-1800 elo can work out if the team assignment is balanced. a duel with 1400 elo vs 1800 elo is disaster)

3) show people the match info (mode/map/teams/skillratings). make everyone accept the match before throwing people into a server. if there's someone who doesn't accept it, repeat step 2

4) play quake with a mercy limit that lets the losing team vote to abandon the match before the full duration (should be rare, but lets people avoid bans)

5) ban people who quit ungracefully from matchmaking for the next 10 days

Has anyone access to raw usage stats of the game? I guess the qlranks elo bot crawls all matches, so he knows when/how many/where people play. A database dump of the last months would be very helpful to check a test implementation of matchmaking against real world usage patterns.

I play tdmpickup mostly. From what I hear the channel is quite active compared to other teammode matchmaking queues. Sometimes the games there are unbalanced, but most of the time the bot does a decent job. Wait times differ from 10 minutes average in peak times (a new game starts almost immediately after previous finished, game duration 20 min) to 1+ h during the work day to >6+ h during euopean night.

The span of skill in this player pool is infinite(i.e. >1800). From the tdm allstar players to 1200 Elo scrubs with -40 net frags everyone plays in the same pool with o.k. results. The bot does not use the qlranks elo system but a handcrafted tdm-rating, that is way more accurate. So, to get decent results I guess a mode-specific rating algorithm has to be done. (Elo is an o.k. starting point for this maybe?)

For team-modes with 4+ players per team I don't see a huge problem, given a working individual rating-system. Of course, the waiting times get too long if the player-base is below a certain threshhold per area. That's something that would have to be accounted for by looking at the usage stats beforehand!

For duel, the restrictions on skill-gap are much tighter but at the same time the player base is larger and you need a factor of 4-5 less players to get a match going. Hard to tell without specific player numbers at a given time of day, if this would produce o.k.-ish waiting times.

But if the player numbers are low, one could always show a list of duel servers to spectate will remaining in the queue and this way not let people quit by boredom. Just let them spec a very high skilled duell (or one of their skill) while getting send to a match immediately when a possible opponent joins the system -> Everyone wins!

"but most of the time the bot does a decent job"

At failing. Otherwise we don't play the same pickups. When I KNOW that it WILL BE mercylimit even before start. And it becomes just that. : D

Still better than tier-system but far from decent.

Ok, let's put aside the 'under the hood' details for now.

What would the ideal scenario look like from a User Interface standpoint?

Would skill elements be shown on the scoreboard, or in the match summary, profile, or external list?

Big red "Play" button on top. Selecting gametype afterwards, the way it works in LoL.

Showing or not showing skill is a separate issue. If they want to carebear players (which they probably should), they should either not show it, or show it to friends only. Even if it's shown, it should probably be only in division (Bronze to GM or whatever), not in specific rank. Flaming teammates for 5 less elo can get rather toxic.

Just give me filters so I can show only FFA + CA, only from DE, PL, not empty, not full, 2 minutes into the game or less, ping < 70, no insta or other mods.
And then show meh qlranks ELO.
Without this clusterfuck of click left look right look left look right look left.
I will matchmake myself
DansGame

tis a million times

Isn't there too small a player pool to do effective match making?

haha. :]

not after the steam launch, at least not for a while.

Short synopsis

CA

smurf slammer joins server. rails at will in designated corner. Does not accumulate anything but kills. Ends up 2nd or 3rd on board > remains in tier 3. If he moves up> makes new account

rinse and repeat

++++++++++

TDM

smurf slammer joins tdm server, kills his own teammates and himself, then proceeds to kill everything else in sight with 60% rail > moves up to tier 4 but so does everyone else because of low pop

rinse and repeat

++++++++++

CA

smurf slammer joins ca server. destroys everyone on opposite team with 50% lg. moves up a tier > makes new account

rinse and repeat

++++++++++

tldr: fail system didn't address loopholes. One sided convo about matchmaking instead of focusing on fixing the exploits these "smurf slammers" use to bypass the system and the fact that the most known of that group has yet to be banned into oblivion, still sitting on servers destroying the new players entering the game.

reason = too many are friends with the devs/elites in this community. That's why you don't see this addressed here or elsewhere.

Play more TDM pickup. Such faggots doesn't exist there.

first para was meant for CTF

too much elitism in pickups. it's only meant for tier 4 players

Proper friendlist.


Whine solved.

In terms of User Interface, maybe Frozen Synapse's "traffic light signal" system could be worth some investigation: http://fs.error420.com/wiki/Matchmaking

More than anything, tiers are incomprehensible to me for duel matchmaking.

4 Tiers? There are arguably 2800 or so 'tiers' in qlranks, and even that has problems with the confidence of that ratings between players who have never played each other or who each play with a separate group of people (doomsday being 2400 elo when he really wasn't even 2000 tbh).

The other thing that really gets me is: the problem with most matchmaking systems after the basic premise/formula (which is flawed in QL, w/e it is) is deduced is attaining the data. A game such as QL which records demos is just sodding perfect for accumulating more data.

I mean, imagine integrating the weapon damage taken for each players life; how much damage each player did when they had the lower stack/higher stack; avg player movement speed; viewpos of killer; viewpos kill-ee and on and on and on....

A script which compiles this data on the fly or from a demo (that could be uploaded to a main server and deleted after if need be) could deduce if a player wins through high levels of aim, from repeatedly outdamaging in the same situations game after game, whether they are any good at item timing/rotating, movement; their strength out of control/in control and much more.

I am by no means claiming this should have been put into the game, just the potential our beloved game has for this, the what if, and the what could have been :'(''

SteamWorks.

eot?

Matchmaking for tier 4 is fine imo, at least in australia.

There isn't enough players to make a tier 5, if there was a tier 5 it would only be a handful of people in it for duel. Also, the 'top' players in australia usually just organise games between themselves anyway.

I think a distinction should be made between duel and team games.
1 on 1 duel is the easiest case imaginable to do matchmaking and you don't need anything more than pure (ELO)rating. (ELO was designed for 1on1. *)

For teamgames & tier 4: just splitting tier 4 in the middle would already be a big improvement imo.
Which would result in more than just "a handful" of players. But they had tier5 and dropped it, presumably for good reasons.

* On chess servers (which only use ELO) if you wanna play you throw out a challenge: you specify time limit and rating range. There's a list with all outstanding challenges. Either you wait till somebody accepts your challenge or pick a challenge from the list. There are so many people playing on those servers (playchess, icc for example) that you never have to wait for more than a few seconds. It is often hard picking a challenge from the list because it changes so fast. (It's gone by the time you manage to click on it.)

ecs made some posts about this couple of weeks back on the official forums, might clear up some confusion.

Id don't use anything similar to Elo at the moment. They provide a tracker now to qlranks.com to make it easier for them to operate. The main problem is in tier4 as ecs explains in his posts. It would be easy to create a match maker. It would be done through scripting (hook/extraQL) or simply a replacement client. It will come soon no doubt.

http://www.quakelive.com/forum/showthread.php...post300319

http://www.quakelive.com/forum/showthread.php...post300387

I hear this a lot, but I'm curious as to how exactly this would manifest itself for the end user in Quake Live.

That you end up in totally uneven teams, that's how.

The only way for this system to be satisfactory is to remove arbitrary automatic matchmaking.

Users should see the QLRanks elos ingame in warmup - and rearrange themselves properly.
This would also help the server admins to view and assess the teams properly - ant kick stat whores and stackers if needs be.

Don't worry, the community would quickly put some wannabe slummers in their place.

Concerning duel and matchmaking:

Too small of a player-base is not an adequate excuse to not implement matchmaking into Quake Live. Whether you have a player-base of 10 or 1,000 in a particular region matchmaking would still be beneficial.

We currently idle in a game until someone else connects, or we attempt to arrange a game via our friends-list. The problem with idling is that you get kicked for idling (some players don't even look for idlers, opting only to play arranged matches), and the problem with arranging matches via a friends-list is that not all available-players within a region are on our friends-list. We all have incomplete lists.

It takes time to build an incomplete friends-list. Over time players have to bump into other players that are willing to accept a friend-request. Gradually we've built an adequate list of players, but new-players have an empty friends-list. Therefore they can't arrange matches. Their only option is to idle in matches that will kick them for idling.

This gives the illusion of a "dead" game. Empty duel matches as far as the eye can see . . . because older players aren't fond of idling: it's ineffective and they already have an established friends-list. What's more some matches are hidden due to the filter being used, or the skill-tier of the match. Further adding to the illusion.

(Is it any surprise CA sustains the highest population? New players are always bound to see an active CA match.)

The point is: the community has no way of gauging a region's persuasion to duel at the onset. They have to build a friends-list, which takes time, dedication, and is always incomplete. It's ineffective, and a hindrance to growth. A matchmaking system, regardless of the player-base population would offer a tool to gauge a region's persuasion to duel, and it would help foster growth of the duel player-base in that it would retain players who previously left because they couldn't find a match.

In other words: it would be helpful to the existing population, and to the future population, regardless of current population.

I'm going to duel in a region that claims to have a small population of players. I have to manually investigate every player on my incomplete friends-list with a nice, polite message requesting a duel (and I add everyone who I encounter). Sometimes people don't respond--either because they're AFK, don't want to duel, or miss my message. I have no way of judging which of the three scenarios my message encountered . . . so I just hope someone responds. If someone does we can duel, and I've been successful in my mission. Sometimes no one responds, and I try idling in an empty match, but no one joins. The regions persuasion to duel at this particular time is negative, and I've just wasted half an hour.

Don't you see? We have a matchmaking system already! It's us. Only it's manual (requiring player effort), and exclusive (only available to players with a built-up friends-list); therefore incomplete and ineffective. Still, it works, doesn't it? I am able to find a duel partner most days, in a region that claims to have a "small" population.

An automated, systematic matchmaking system would be complete and more effective than what we have now. No one can claim it wouldn't work for duel, because we already have a broken, exclusive matchmaking system that does work ineffectually.

I forgot to mention that a community-made matchmaking system wouldn't be helpful, as it doesn't address the issue of exclusivity. New and the majority of existing players ( I presume) don't know what ESR is, and if they don't know what ESR is how are they going to use a community-made matchmaking system that will (presumably) be posted on ESR? It will trickle down from players who share it, but it doesn't address the problem.

The "Play Now" button that was recently implemented doesn't address any problem that the server browser doesn't. I'll argue that it's potentially detrimental, because when the influx of new players come they may be confused at being told they can't join a match because it's for premium only players, or that their skill-tier is too high.

Lastly, I don't support a automatic, systemic matchmaking system for "even" or "fair" matches. I don't believe the dificulty is scaring new players away, but the inability to find matches is. That's why I want an automatic, systematic matchmaking system for duel: to easily find matches. The quality of the matches is irrelevant at the present time, but as the player-base grows it will become more important. Thankfully, with a larger player-base the quality of matches could be improved.

As I described above for chess, a "challenge" list should be easy to implement no?

If you want to play you issue a "challenge"; it includes your rating, preferred rating range for opponent, preferred server location, maybe preferred map.

All "challenges" are collected in a list that everybody can see.
Either you wait for somebody to accept your challenge or you pick one yourself from the list, meaning contacting them on chat so you can arrange a game. (Obviously you can just pick a challenge from the list & start playing without bothering to issue a challenge yourself.)

For duel it could be an option.
For the other modes where more than 2 players play, not so much.

But I think that in QL implementing it would be kinda shitty for new players as in general new players are inserted into rankings as average players... So most new players would quicky decline in rank, which is kinda frustrating.
Ofcourse you can also do a system a bit more like the system used by KGS (client for the game of go) where you play a few ranking game and you get put on a 'rank' from 1-30 according to that (where 30 is bad and 1 is awesome)

Don't understand your comment about new players. You start with 1200 (or whatever is appropriate) and work your way up. * (If QL does it differently for duel then they're doing it wrong.)

Anyway, this is not an automatic system. It's just a list and you pick and choose yourself. (So if you don't want to play a low rated player, don't accept his challenge.)

*{Also Elo has system built in (k-factor) to fine-tune the rating calculation a bit. For example beginners have provisional rating, meaning the k-factor is bigger, k-factor is lowered as you play more games. (Big k-factor means bigger win/loss per game, more volatile)}

the comment on the new players is based upon the 'idea' that most players in quake have been around (in quake) for ages. And therefor the "average player" (1200 elo) is actually much better than by far most new players.

Aka if you have a scene were "bad people drop out" and "skilled people stay in" the average slowly shifts in skill while it remains almost stagnant in elo.

Also the average level upon the start of elo was massively higih. So the average is skewed even more

>> the "average player" (1200 elo)

In my example the 1200 is the starting rating for a first time dueler.

(And this does not need to equal the average rating of the whole player-base. Actually: it shouldn't be. A new player would realistically be much weaker than the community average.

But you know all this (I assume). I think you were misunderstanding me somewhere..

ehhh in Elo rankings you let all people start on the average.
(or rather you let everyone start on the same number and that is the average because elo points gained by one person = elo points lost by the other)
At least that's the standard elo model.

Ofcourse you can differ by changing it in such a way as you say. But then it's not officially elo anymore (not that I mind).

>> ehhh in Elo rankings you let all people start on the average.
(or rather you let everyone start on the same number and that is the average ...

I don't follow you.
Also I don't think that is how Elo rating systems are usually implemented.

(Although you could do it like that; the starting rating of a new player really is not that important. Whatever it is, the new player will quickly arrive at the rating he is supposed to be at.)

In chess the worldwide average is somewhere between 1900 and 2000. I don't know of a single federation or community that starts people on a rating in that range, it's always much lower.

To really get into a discussion about rating systems I would have to start reading up on Elo (or similar) rating systems, and I don't care enough about this to do so.

More importantly: my initial comment in this (sub-)thread described a list of challenges that can be used to find somebody to play with. That has nothing to do with how rating systems work.
Moreover, the Elo rating system (or how to use it) is well established theory and proven to work, I don't see the point in discussing it any further.

If everyone has the same K factor, then the average will stay the same as the starting value. N-N=0 This is how QLRanks works. Many implementations give their new players a higher K factor though.

I'm not sure what you mean with your "worldwide average is somewhere between 1900 and 2000"... do you have any source? We would need to know exactly what part of the population is considered (I suspect it's indeed worldwide... but among players above a certain level)and, of course, what the starting value of this system is.

es, you may be right. This is where I was coming from:

"The average value of the ratings is approximately 1900. The rating of the player in the middle of the rating table is 1924. Players with a rating greater than 2300 are in the top 5% of all players, and players with a rating greater than 2450 are in the top 1%."
from http://en.chessbase.com/post/visual-presentat...ss-ratings

The United States Chess Federation gives new players an arbitrary rating of 1300 if they have no other information to guess their strength.

"If an unrated player does not provide a birth date, but is inferred to be an adult (e.g., through an appropriate USCF membership type), then the initial rating is set to be 1300"
from: http://www.glicko.net/ratings/rating.system.pdf

The above 1900 average is for FIDE ratings though, which sets the bar rather high.
I've been searching for other averages but can only find much lower numbers.
So maybe the average for a national federation like USCF really is only around 1300, I suppose it's possible. This blows my mind though, in chess 1300 is somebody who just learned the rules, at least in my experience. I guess there are a lot of newbies.

In any case; to come back to the starting value for a new player: 1200 seems to work for qlranks, what are we complaining about?

what are we complaining about?

In my case, I was mostly startled by this claim :

In my example the 1200 is the starting rating for a first time dueler.

(And this does not need to equal the average rating of the whole player-base. Actually: it shouldn't be. A new player would realistically be much weaker than the community average.

Because in the simplest form of Elo's rating system the starting value is the average value. But as you said, a real newbie is typically overestimated at the average/starting value and will quickly drop.

I kinda said that a real newbie is overestimated at the average....

Anyway thx for the update about different K's in the start of games, I didn't know that it was actually part of elo. I thought it was some derivative of it.

Anyway2 I don't think that anonymous gets the difference between an instance of a ranking algorithm and the ranking algorithm.
It sounds as if he thinks that within all elo systems the average would be the same as it is in chess. (around 1900)

There's no "official Elo ranking system". Well I guess you could consider what you can read in Elo's papers the "official" system. But really who's to say what's Elo and what's not? I'd say small changes like dynamic K values and using a logical curve instead of a normal one still qualify for Elo.

To be honest I haven't read Elo's papers so I wouldn't be able to tell exactly what he advocates.

I think I've read too many scientific papers in my life already... Silly questions on messed up details spring to mind constantly and it's very useless :D

Anyway the "official Elo ranking system" should be the one proposed in mr. Elo's first paper on the subject. Naming of algorithmic stuff usually works like that. Any alteration on it is usually named different.
But I'm starting to get boring on details and I have better stuff to do myself so...

It sounds as if he thinks that within all elo systems the average would be the same as it is in chess. (around 1900)

That does not make any sense. I may sound stupid to you, but I'm not that stupid. One rating pool can be totally different than another, even within the same game if they never mix.

stupid is something completely different that unfamilliar with.
A lot of terminology makes me go "wtf do they mean by that" and I'm not stupd either. So I look up what is meant by the term and then I understand what they mean ;)

im not stupd either

I didn't say you were.

stupd

>>> Because in the simplest form of Elo's rating system the starting value is the average value. But as you said, a real newbie is typically overestimated at the average/starting value and will quickly drop.

As I already hinted at, when you play against somebody who just learned the rules and then somebody else comes along and says: oh yeah btw, that newbie; that's the average of the playerbase... allow me to be skeptical. In my experience a newbie typically will quickly rise in rating.

But yes, if the formula is exactly the same for everybody (so also the same k-factor, no provisional ratings etc) then yes, the starting value is the average. But in my experience with rating systems this is not the case, something else is going on to make the average higher than the starting value. I assume this is because of variable k-factors (big for newbie, small for experienced players) and provisional ratings (for newbies).

I have a follow up question.
Most chess rating systems that I know use a variety of k-factors and use provisional ratings for beginners. Am I right in assuming that this will lead to a higher average than the starting value? It's almost impossible for me to otherwise accept that the average can be that low. Even for QLRanks an average of 1200 seems way too low.

Yes, using such techniques can lead to a different average. For instance, if you give your old player a K of N and your new players a K of 2N, then if the new player wins, the amount of points lost by the old player will be less than the amount of points won by the new player. Which means that there are points introduced in the system without adding new players at the same time, which rises the average of points per players. Of course the opposite would happen if the new player were to lose. I don't know how well it would balance in the real world.

As for QLRanks... look at this the section "Top 50 Countries by No. of Players" :
http://www.qlranks.com/duel/

In general "the average player" would mean that half of the players lose to him and half of the players win. So it should even out.
However that is excluding the idea that players are sometimes tempted to make new accounts when their elo is nearing one of the extremes (high / low).
I would figure that most "average players" who remain around the elo average would not have the need to create new accounts etc.

Well there's a big difference between the average Elo and a median player.

I somehow doubt that. When the algorithm is set up like this they should remain very close to eachother.
Assuming that everyone starts on the same elo and that the K stuff equals out over time or is not used.

The algorithm assumes it's a perfect normal distribution but that may not be the case in reality. I suspect there are more players below 1200 than there are above it.

It's pretty much irrelevant anyway... I'm not even sure why we're discussing that.

hehe ok let's not discuss it further.
I can see your point. I'm not entirely convinced that you're right about more players below 1200 than above, but you're definitely right that it doesn't matter :D

How about this:
http://www.chess.com/blog/DrawMaster/uscf-rat...is-bimodal

I wonder if QL's skill distribution is bimodal too. We know that the global average is 1200. We know that the vast majority of QL players are terrible and not competitive minded at all. Also, take a look at the skill distribution I computed for targetquake's league a few months ago :

https://dl.dropboxusercontent.com/u/74076095/...league.png

that would indicate that I'm right. Thx for the link.
(did you read it?)

>> Yes, using such techniques can lead to a different average.

Ok. So I was right after all, the average can be higher than the start value, it just depends on the implementation.

>> As for QLRanks... look at this the section "Top 50 Countries by No. of Players" :
http://www.qlranks.com/duel/

Holy crap! Didn't expect that.

I thought the average in qlranks was 1500 myself, not entirely sure though.

>> I thought the average in qlranks was 1500 myself, not entirely sure though.

There you go. Even in cases where the average DOES equal the start value appearances can be deceiving.

Seems like I was never near the average then....
First time I checked my ql-rank I was around 1600 elo and they had just started measureing... Kinda assumed I got there from starting on 1400-1500 and increasing a bit.
It does kinda show that I'm right about how the algorithm works though... which is stuff that I should understand ;)

http://www.qlranks.com/duel/player/weird
Page 285
You were at 1200 on your first match. You lost it against "reload" and lost 15 points.

damn that reload taking my elo points grrrrrr :(
I never really paid attention to it hehe :D

whatever algorithm they use for determining skill, it's certainly NOT ELO (or at least not unmodified) and it's horrible:
I'm on ~1100 ELO. Yesterday when I was searching for a game in the server browser QL's skill indicator thingie matched my qlranks ELO mostly fine, i.e. players with slightly higher or lower ELO were considered "Your Skill", those with way lower ELO were considered lower than me etc. Then I played 3 duels against a 1400 ELO guy, lost all of them not close at all and then called it a night.

Today I log in and and a 850-ELO guy sitting alone in a server is now considered "more challenging" for me (the orange one). gg.

Not this again .. :/

QL absolutely does use Elo for duel.

See http://www.esreality.com/post/2641476/ql-match-making/#pid2642244 from this very page even.

Maybe this 850 player joined a much higher server but the other players quit.

Or maybe simply this 850 Elo players had just won a few matches again someone considered better than him. QL's Elo implementation has a very high volatility. It doesn't take much to change the skill estimation greatly.

But it would imply that the implementation is kinda wonky. Can it really be that hard to fix that (at least for duel)?

Well I pointed stuff out before on QL being too volatile (if it uses elo).
Back when I was extremely active and competing for high places in tournaments... often after just losing 2 ~ 3 matches I would suddenly see several servers which were "extremely difficult" with players om them who I never lost to before.

So I'm not sure what "used to be wrong" but theres definitely something wrong if my elo on ql ranks tells me I'm in the top100, while in QL there are several servers "extremely difficult".
(matched, difficult, very difficult, extremely difficult... That's 3 "levels" apart... seeing how there only are 4 levels in QL that would say a lot innit?)

(matched, difficult, very difficult, extremely difficult... That's 3 "levels" apart... seeing how there only are 4 levels in QL that would say a lot innit?)

The 4 "levels" do not apply to duel in this situation.

So it's "within your own level" that the subdistribution is made between matched / difficult / etc?

There's simply no concept of "tier" for matched/easier/moredifficut/etc. It's only based on Elo difference between you and the (estimated) best player in the server.

right so when I was in the top50 in QL ranks... after losing 2 matches, suddenly people with 1600 elo were "very difficult" while before those 2 matches the same server showed my "below your skill".
So according to the ID implementation I would've lost around 500 elo in 2 games... I would be inclined to say that they made some error.

Yes it's quite ridiculous.

agreed

Can't we all hold hands, sing songs, and agree that sponge is just a super, ultra, shitty coder with an almost hilarious lack of clue as to what the fuck he's doing?

Yay!

QL seems to think my skill is all over the place, too. The other day a guy with a >600 ELO advantage was grey for me, and there are cases at the other extreme where players around my level can't even join my server because I'm too shit.