80 replies to this topic

[Nightbird]

Kroete, on 26 November 2020 - 10:58 AM, said:

That all would be true, if we would not have groups and would have even weights in a match.
Until then all the talk doesnt matter much, as still having 5 tiers, even if we coulnt support tier 1 and 5 with enough players without opening the valves.

Let me refer you to these threads, created back when PGI was soliciting ideas for updating the Matchmaker. It shows how we were able to accurately model how well a Matchmaker would perform. Shows that math should be involved in the discussion.

https://mwomercs.com...cy-with-graphs/

https://mwomercs.com...psr-comparison/

[Rustyhammer]

Nightbird, on 24 November 2020 - 06:50 AM, said:

You claim all of that 10% group that left before finishing 10 matches didn't do so because of the Matchmaker but some other reason. Based on what? 25%, 50%, 75% of them could have had 5 matches and said enough stomps, and quit, what's the support for your claim? I have no data.

Most of your data came from the old PSR time, when new players were placed in T5.
I don't think there were regular bad stomps in T4/T5 matches. If anything, I'd say NPE was much better then in today's soup queue.

I could be wrong though, and excessive player bleeding from both lowest and highest WLR bracket is mostly caused by the bad Matchmaker. Maybe the impact of low WLR newbies quitting MWO after a few matches is insignificant, as well as high WLR players retention rate is not skewed by things like T1 smurfs reaching T3 and creating a new T5 account to club seals.

Would it be possible with stats you collected to make a graph like

X axis: number of matches played (or number of seasons played)
Y axis: number of players left

for each of WLR brackets? If you going to do that, could you please also exclude soup queue seasons, as NPE experience is wastly different now comparing to days of old PSR.

[Zordicron]

Go back to 8v8.

Shortens wait times. Less players needed.
Shortens wait times. 2 of each weight class vs 3 for 12v12
Valves should not have to open as far, or as fast, because wait time is shortened.

Helps new player XP
Get into matches faster, get into better matched teams and matches.
Reduces overall firepower on the battlefield. Less gankfest deathball. Less NASCAR.

Yes, there are other things to also do with matchmaker, I don't dismiss everything in this thread, in fact it is all spot on. But there is more to it then one single thing when it comes to player retention. MWO is a niche in the FPS games market, slower, tactical, thoughtful play vs sprint shoot die respawn. Newbs need to be able to make a mistake without being insta-cored and run over. Better skill match ups helps, but even 12 newbs have enough firepower to punish someone else's mistake in short and fatal order. We need to reduce firepower on the field.

[VonBruinwald]

Zordicron, on 26 November 2020 - 05:54 PM, said:

Go back to 8v8.

Shortens wait times. Less players needed.
Shortens wait times. 2 of each weight class vs 3 for 12v12
Valves should not have to open as far, or as fast, because wait time is shortened.

Problem with this is they'll have to separate group queue again (which didn't work even after they dropped it to 8v8), limit groups to two or risk watching four-mans drive the last of the pugs away.

Now the first two don't cater to groups so PGI won't do them, the last will kill MWO. Which I suspect is something Russ is secretly going for anyway.

Edited by VonBruinwald, 26 November 2020 - 06:34 PM.

[MrMadguy]

VonBruinwald, on 26 November 2020 - 09:27 AM, said:

There's a flaw there, you are assuming good players will pull the teams performance up more than the bad players will drop it. With a functioning matchmaker the bad players will pull the team down an equal amount to what the good players raise it relative to the opposite team.

How can bad players pull your match down? Team kill you? It can happen only when you need support, but you don't have it because of bad players on your team. But this implies, that teams are imbalanced right from the beginning. But what if they are? You don't need support in this situation. You just play your game and bad players just die early without any meaningful contribution towards victory.

Do you imply, that I'm bad player? I'm not bad. I just need some conditions to perform good - when enemy players don't overrun me. And they don't, if they have skill, that is close enough to mine. My yesterday's 3 kills and 800+ DMG match proves, that I can do well. Problem is - it happens very rarely nowdays. Overall it's just purely random accident, not pattern, as it should be.

I survived with armor stripped from all parts, except legs, despite of last enemy players trying to focus me. It's deadly situation for IS 'Mech, as it's enough for it to get one of STs or CT hit in order to be one-shot. And when I play against pro players, it's enough for them to strip armor from just one part of my 'Mech to one-shot me almost immediately. Simple fact, that I survived this match, shows, that guys, who were playing against me, had pretty bad aiming skills. I.e. something on a par with "I'm good, if at least I hit enemy silhouette".

As I've already said, decent match is combination of weak enemies and weak teammates. If enemies are weak, you kill them. But your team also shouldn't kill them too fast. Otherwise they'll leave you with nothing. This clearly shows, that team balance itself isn't enough for good matchmaking. Making teams even - is mandatory requirement. And may be WLR MM can work towards proper direction, but it doesn't have enough momentum to achieve this goal. If conditions are bad, it can stuck in dead loop of self-induced problems and fail to get out of it.

Overall WLR matchmaker doesn't see any bad in following situations:
1) Balanced, but uneven teams
2) 12-0 stomps interleaved 50/50 with 0-12 ones.

Both situations are bad and proper MM should avoid them. WLR one doesn't.

Edited by MrMadguy, 26 November 2020 - 10:54 PM.

[Nightbird]

Rustyhammer, on 26 November 2020 - 12:23 PM, said:

Would it be possible with stats you collected to make a graph like

X axis: number of matches played (or number of seasons played)
Y axis: number of players left

I could graph this but it wouldn't be useful because both are outward symptoms of the happiness of players. Someone who is happy will play more games and have high retention, someone unhappy will play fewer games and have lower retention. To use an analogy, it would be like graphing how often people use Facebook/Twitter against how useful they rate these websites to be. It's useless self-reinforcing correlation.

Graphing WLR against Retention is useful because WLR is a controllable consequence of the Matchmaker rather than an indication of the happiness of a player, and retention is useful as a metric to predict revenue. I could graph WLR against average matches played but that clearly wouldn't be as interpretable as retention.

Edited by Nightbird, 26 November 2020 - 09:49 PM.

[Kroete]

Nightbird, on 26 November 2020 - 11:14 AM, said:

Let me refer you to these threads, created back when PGI was soliciting ideas for updating the Matchmaker. It shows how we were able to accurately model how well a Matchmaker would perform. Shows that math should be involved in the discussion.

Your models seems ok, under ideal conditions,
if we would have even weights and no groups and no tiers.

If you want to model it accurately then you also need to count for groupperformance:
4 soloplayers each with a psr of x is not the same as the same players playing as coordinated group (in most cases).
Even weightdistribution is also needed or the difference must be counted in the model:
If you have even teams, the team with 200 tons more will mostly win.
And tiers need to be gone, all the models you linked work with psr not tiers.

So we need to do some basic points until you model will work not only on the "paper":
1. Remove tiers, there are not enough players for tiers and open valves will throw the balance all over.
2. Weightbalanced matches, find a point whats the max difference is for not unbalancing matches.
3. Groups, how should they be weighted against solos for balanced matches (removing would be easier)?

Edited by Kroete, 27 November 2020 - 04:49 AM.

[VonBruinwald]

Kroete, on 27 November 2020 - 03:58 AM, said:

3. Groups, how should they be weighted against solos for balanced matches (removing would be easier).

Group PSR = (Average of the top 2 players in the group) x (Number of players in group) x 1.5

The 1.5 value can be tweaked if groups are being over/under-weighted relative to pugs.

[MrMadguy]

Kroete, on 27 November 2020 - 03:58 AM, said:

2. Weightbalanced matches, find a point whats the max difference is for not unbalancing matches.

I've got tired of constantly fixing this misconceptions for you. Skill >> weight. My performance in Assaults is lower, than in Heavies, except may be some OP variants, like LRM/ATM boat with ECM, TAG and Gauss support, that is good in all possible situations. Assaults are big and slow. They have to expose themselves, if they want to do damage. They're actually extremely easy so kill. Especially IS ones, that suffer from ST death penalty.

[Kroete]

MrMadguy, on 27 November 2020 - 04:59 AM, said:

I've got tired of constantly fixing this misconceptions for you. Skill >> weight. My performance in Assaults is lower, than in Heavies, except may be some OP variants, like LRM/ATM boat with ECM, TAG and Gauss support, that is good in all possible situations. Assaults are big and slow. They have to expose themselves, if they want to do damage. They're actually extremely easy so kill. Especially IS ones, that suffer from ST death penalty.

Its not about mechweights and your performance, its about the weightdifference of the teams in a match,

[MrMadguy]

Kroete, on 27 November 2020 - 05:09 AM, said:

Its not about mechweights and your performance, its about the weightdifference of the teams in a match,

And at the end you imply, that when I play Assault, my skill should be weighted higher, than when I play Heavy. No? So again, game would expect me to do more, than I can provide, as it was doing back in exp bar PSR era.

[Nightbird]

Kroete, on 27 November 2020 - 03:58 AM, said:

Your models seems ok, under ideal conditions,
if we would have even weights and no groups and no tiers.

If you want to model it accurately then you also need to count for groupperformance:
4 soloplayers each with a psr of x is not the same as the same players playing as coordinated group (in most cases).
Even weightdistribution is also needed or the difference must be counted in the model:
If you have even teams, the team with 200 tons more will mostly win.
And tiers need to be gone, all the models you linked work with psr not tiers.

So we need to do some basic points until you model will work not only on the "paper":
1. Remove tiers, there are not enough players for tiers and open valves will throw the balance all over.
2. Weightbalanced matches, find a point whats the max difference is for not unbalancing matches.
3. Groups, how should they be weighted against solos for balanced matches (removing would be easier)?

Here's no data for your claim that 200 more tons win more. It could be true, or false. If PGI released data for every match (team members, mech, performance, etc), I could analyze how much of an effect tonnage has. It's true that heavier mechs earn slightly more Matchscore on average than lighter mechs, but Matchscore is not strongly correlated with wins.

[Knight Captain Morgan]

Zordicron, on 26 November 2020 - 05:54 PM, said:

Go back to 8v8.

Shortens wait times. Less players needed.
Shortens wait times. 2 of each weight class vs 3 for 12v12
Valves should not have to open as far, or as fast, because wait time is shortened.

Helps new player XP
Get into matches faster, get into better matched teams and matches.
Reduces overall firepower on the battlefield. Less gankfest deathball. Less NASCAR.

Just take it all the way down to 2v2. Maybe even hold the matches in cool, smaller Solaris style arenas...

[Anomalocaris]

VonBruinwald, on 27 November 2020 - 04:56 AM, said:

Group PSR = (Average of the top 2 players in the group) x (Number of players in group) x 1.5

The 1.5 value can be tweaked if groups are being over/under-weighted relative to pugs.

The only problem with that is below average groups have a negative impact on team performance and grouping a bunch of sub-par players together tends to make them even worse as a whole.

You probably need some sort of exponent to be applied to the difference of the group PSR from average.

Or we just do W/L MM

[Nightbird]

Anomalocaris, on 27 November 2020 - 12:47 PM, said:

The only problem with that is below average groups have a negative impact on team performance and grouping a bunch of sub-par players together tends to make them even worse as a whole.

You probably need some sort of exponent to be applied to the difference of the group PSR from average.

Or we just do W/L MM

W/L covers it, if on average. People benefiting from groups will have higher W/L and vice versa.

A better way that almost no one here would understand would be to use a Bayesian approach. The solo drop W/L would the "prior", and for each other member in the group, a group W/L when paired with that person would be recorded and used as the "likelihood". Finally, a player's "posterior" W/L would be calculated. This final calculated value would represent the player's skill as a member of that specific group. This calculation method would take care of all groups in queue and force 95% of the pop into the 0.9-1.1 WLR region, it also adjusts to when a player plays solo or changes group members, without unfairly over or under-evaluating the player, unlike just having one dumb W/L.

Edited by Nightbird, 27 November 2020 - 01:09 PM.

[Rustyhammer]

Nightbird, on 26 November 2020 - 09:47 PM, said:

I could graph this but it wouldn't be useful because both are outward symptoms of the happiness of players. Someone who is happy will play more games and have high retention, someone unhappy will play fewer games and have lower retention.

That's why I said build this graph per WLR bracket, wouldn't it help to sanitize data a bit?
Again, my feeling is that majority of new players who played extremely low # of matches also have low WLR and their stats driving retention rate for WLR < 0.7 down.


Anyway, I'm feeling this discussion is turning into Surn 2.0 territory so I'm going to stop.
You're using data collected post several major player exoduses and yet claiming that '50 out of every 100 players that left this game has done so because of the Matchmaker'. You don't want to investigate if there is any significant pollution of retention rate for low WLR players caused by new players trying and dropping this game. Then why asking for comments?

P.S.
I'd love to see a similar discussion of how establishing and forcing people into having the optimal number of dentist visits per year should improve teeth retention rate for everyone. And how it will help to decrease teeth loss for people < 10 years old.

[Nightbird]

Rustyhammer, on 27 November 2020 - 03:27 PM, said:

Again, my feeling is that majority of new players who played extremely low # of matches also have low WLR and their stats driving retention rate for WLR < 0.7 down.

Again, why shouldn't new players have a 1 WLR while they are learning?

Rustyhammer, on 27 November 2020 - 03:27 PM, said:

You're using data collected post several major player exoduses and yet claiming that '50 out of every 100 players that left this game has done so because of the Matchmaker'.

Because even during those exoduses, it clearly shows that the green WLR bracket players are retained at twice the rate of other players?

Rustyhammer, on 27 November 2020 - 03:27 PM, said:

I'd love to see a similar discussion of how establishing and forcing people into having the optimal number of dentist visits per year should improve teeth retention rate for everyone. And how it will help to decrease teeth loss for people < 10 years old.

A more apt comparison is lifting weights at the gym. To much weight and you can't do 1 rep and get frustrated. To little and you feel no gains. Need it to be just right.

[Mech Walesa]

@nightbird
in starcraft the MM is set to 50/50 with 5 tiers and 6th tier (grandmaster) with 200 places.

Edited by Mech Walesa, 29 November 2020 - 01:08 AM.

[The6thMessenger]

Nightbird, on 27 November 2020 - 11:48 AM, said:

Here's no data for your claim that 200 more tons win more. It could be true, or false. If PGI released data for every match (team members, mech, performance, etc), I could analyze how much of an effect tonnage has. It's true that heavier mechs earn slightly more Matchscore on average than lighter mechs, but Matchscore is not strongly correlated with wins.

The dude seems to be asking for it.

Besides, maybe proper analysis will probably put his concerns to rest.

Edited by The6thMessenger, 29 November 2020 - 01:28 AM.

[Nightbird]

I think you missed the part "If PGI released data...", can't analyze data that's not available.