357 replies to this topic

[Capt Deadpool]

C64 Warrior, on 19 June 2020 - 03:19 PM, said:

Because your personal skill has nothing to do with winning or loosing (there are 23 other people in any given match that effect win/loss) I vote 1A

No one has ever said this, ever, my dude. Every player from the lowliest potato to a world champ experiences good and bad streaks due to teammates, so luckiness/unluckiness is evenly distributed, and over time, true skill will be reflected in your W/L (if 'true skill' is defined by making winning your primary goal when you play a match... which it should).

[Hellbringer]

hey is blazing sushi open for lunch next door to you PGI guys? been craving that cheapie aburi sushi.... huhem... back on topic

Paul, this looks like a good step forward with the amalgamation. Suggestion is to try it, the more testing you do, the more data you'll collect and you can finetune the results afterwards. First step is to reset that PSR pronto, then we can move ahead from there and finetune as needed.

[C64 Warrior]

Capt Deadpool, on 19 June 2020 - 04:33 PM, said:

No one has ever said this, ever, my dude. Every player from the lowliest potato to a world champ experiences good and bad streaks due to teammates, so luckiness/unluckiness is evenly distributed, and over time, true skill will be reflected in your W/L (if 'true skill' is defined by making winning your primary goal when you play a match... which it should).

But it won't happen if winning is the prime criteria PSR is based on. It becomes a win rate match making system where the more you win the more likely you will be on the team that wins next time and the more you loose the more you will be placed on the loosing team next time. The function of the MM is to rate your PERSON skill and to put you in a match with equally skilled players, to do this you need to measure a persons INDIVIDUAL skill not just their luck in dropping on the winning team game after game. If you are evaluating 1v1 matches then yes winning could be a criteria, but this is 12v12, your personal effect on the outcome of a match is only 1/24, your skill should not be based on the other 23 people in a match more than your own performance.

[50 50]

Nice.

A high performing player might have a bad match and get punished for it, likewise an average player might have a blinder of a match and be rewarded for it. As the PSR range is in thousands of points however, the player would have to have consistently good/bad matches to see the actual Tier change. A single really good, really bad match should not be enough to automatically bump the player up or down a tier unless they were right on the threshold anyway.

Quote

Player Skill Rating (PSR) - A player's skill rating in a range from 0 to 3750 with new players starting at 1500. (This is essentially the players Elo rating)

That's 750 points per tier if it's split evenly.
So, we make sure that the change to PSR based on match score is not so dramatic that players are flipping and flopping between tiers every match.

Having read the above from Paul and a few of the others, I had a read through JayZ's document and believe it to be a fair approach.
Essentially it does take into account the win/loss but you effectively have the same sort of outcome where a player on a losing team who does get a good match score and 'fights valiantly' as he put it, does get a PSR increase. And those on the winning team who have a poor match, can still lose PSR.

It's a bit closer to 1A or 1B but applied to both teams with a small weighting for the win/loss.

Worth reading if you haven't.

Edited by 50 50, 19 June 2020 - 05:37 PM.

[ESC 907]

C64 Warrior, on 19 June 2020 - 04:51 PM, said:

But it won't happen if winning is the prime criteria PSR is based on. It becomes a win rate match making system where the more you win the more likely you will be on the team that wins next time and the more you loose the more you will be placed on the loosing team next time.

Whoa, where is THIS logic coming from? If PSR is based solely on WLR, someone that wins should go up and face tougher opposition until they begin losing. Someone that loses should go down and face opposition that is easier until they begin winning. Eventually, everyone should be playing at a level that keeps their WLR relatively neutral!

[Kamikaze Viking]

WarmasterRaptor, on 19 June 2020 - 09:43 AM, said:

So, in regards of the proposal in the OP,

1A : I Support, removes the groups effects and win/loss, you do what you can do in the chaotic mixed queue.

1B : not zero sum, therefore no support,

2A : Can't support as is.
A player that didn't contribute it's fair share to the win shouldn't be carried by the win of the team.
Therefore negative PSR adjustments should happen even if you win!
Reverse also applies.
If this is changed to consider the above, I would support,

2B : not zero sum, no support.

Please Please take a look at the Simulator that Jay Z build which Proves that ALL versions ARE ZERO SUM.
https://docs.google....dit?usp=sharing

[Cluster Fox]

The tremendous benefit of JayZ's formula is X,Y and C parameters, which allow to tweak the WLR against the personal MS. (It is zero sum too)

This is a very important advantage where the effect of both can be adjusted independently from all the Match Score Kickers (which can be adjusted as well as Paul mentionned).

Chance of the first PSR fix attempt being right without a simulated model is very unlikely. Ability to tweak how much the WLR component is worth for in the PSR is gold.

Edited by Cluster Fox, 19 June 2020 - 08:09 PM.

[General Solo]

Cluster Fox, on 19 June 2020 - 02:30 PM, said:

From those provided, I can only support core 2 options, 2A or 2B with heavy W/L factored in, JayZ or not.
Like, JayZ's 2C with X = 9, Y = 10, C = 1 for instance, which creates a strong shift between win and loss.

I'd like 1A supporters to show data or a simulation that, at least, suggests it will give better matches over time, aka less stomps. Like, IDK, run it for 10000 games. Please provide assumptions and show work.

We've seen data that heavily weighting W/L should do that. I haven't seen long term simulations for the flip side of the argument. Yeah the MM is flawed but PSR and MM work hand in hand.

Decision shouldn't be based on matchscore screenshots and feeling good/bad about a match.

PSR isn't there to make you feel good/bad about PSR, it should give you good matches and ideally reduce stomps. Then you'll feel good cause the matches are better.

My Tier best Tier, change my mind.

I have a question that maybe you or someone can answer as I'm not a stats guy and when I look at JayZ's and Nightbirds beautfull models my eyes go glazly and I no understanding. Luckily its a logic question and not stats/maths.

Anywayz the questions.

Do their models have groups?

In their models whatz the difference between a group and a solo player, both have properties, whats different between them in the model?

How have groups being modeled compared to solo players?

Edited by OZHomerOZ, 19 June 2020 - 08:51 PM.

[General Solo]

ESC 907, on 19 June 2020 - 07:50 PM, said:

Whoa, where is THIS logic coming from? If PSR is based solely on WLR, someone that wins should go up and face tougher opposition until they begin losing. Someone that loses should go down and face opposition that is easier until they begin winning. Eventually, everyone should be playing at a level that keeps their WLR relatively neutral!

Yes but since being solo or grouped in a mixed que has a large influence on win rate that influnce will be carried over into the PSR which rates a single pilot yet being grouped influences this single pilot metric.

[C64 Warrior]

ESC 907, on 19 June 2020 - 07:50 PM, said:

Whoa, where is THIS logic coming from? If PSR is based solely on WLR, someone that wins should go up and face tougher opposition until they begin losing. Someone that loses should go down and face opposition that is easier until they begin winning. Eventually, everyone should be playing at a level that keeps their WLR relatively neutral!

again this is only true if you are playing 1v1 but not in 12v12 matches....you even said yourself winning depends more on the luck of MM placing so how does that become PERSONAL skill? Because possibly as much as 11 people carried you you should get to face better opponents? Really? That is how you would evaluate a persons individual skill? We already have this issue now and how is the quality of the match maker?

[Vindicated]

Since I'm jumping in going to voice my opinion for 2A and 2B because we should encourage more folks to play as a team. If we see folks with W/L far below 0.5 and consistently high match scores, these players might not be good players, they could just happen to be the last one alive by feeding their teammates to the fire. If they were good and lasted that long, they could have done enough to win the game. Even if not one time, then then next time.

OZHomerOZ, on 19 June 2020 - 08:50 PM, said:

In their models whatz the difference between a group and a solo player, both have properties, whats different between them in the model?

How have groups being modeled compared to solo players?

If it wasn't obvious from the people complaining about these groups always winning, these players will have obscenely high W/L, say over 5. They will skyrocket to T1 after a few games and you won't see them again without comparable players on your team.

You know, it's possible for 12 random players to coordinate to some extent and play as team. A lot of the games where this has happened, the game ends something like 12-2, because the other team did not play as a team (notably spread out, came in one at a time, lone assault mech going off to cap points).

C64 Warrior, on 19 June 2020 - 09:28 PM, said:

again this is only true if you are playing 1v1 but not in 12v12 matches....you even said yourself winning depends more on the luck of MM placing so how does that become PERSONAL skill? Because possibly as much as 11 people carried you you should get to face better opponents? Really? That is how you would evaluate a persons individual skill? We already have this issue now and how is the quality of the match maker?

So I can just jump into a game, not do anything, and win consistently? If you're capable of consistently being carried by 11 random people, you're probably doing something right even if it's not reflected in match score (even things such as calling out damaged targets). If not, then it's a fluke and you'll lose a few before you get one winning game.

Remember part of the problem now is that we have players who are T1 and have low W/L and K/D ratios. These players aren't good, just been playing for a long time, and they'd be happy to jump down from T1 because they're matched with players out of their skill level. They've been asking for it.

As far as how does it reflect on personal skill? You said it yourself, it's 12v12 and not 1v1. You're forgetting the ability to coordinate with the team. That is a strong part of player skill, and if you don't think so, why are players complaining about groups? If you can play as a team you have significantly higher chances of winning.

[MrMadguy]

ESC 907, on 19 June 2020 - 07:50 PM, said:

Whoa, where is THIS logic coming from? If PSR is based solely on WLR, someone that wins should go up and face tougher opposition until they begin losing. Someone that loses should go down and face opposition that is easier until they begin winning. Eventually, everyone should be playing at a level that keeps their WLR relatively neutral!

Biggest problem with WLR MM - is that it will be happy with interleaving 12:0 stomps with 0:12 ones. It's self-induced and tends to have positive error feedback. Result of match depends of how good it works and how good it works depends on result of match. So initial mistake in matchmaking can cause wrong match results, that will make further matchmaking even more inconsistent. WLR system won't even notice it. And MS MM doesn't have such problem. If quality of match is bad, i.e. it's uneven, it would be obvious due to uneven players' performances.

Simulators usually don't show this problem. Because they don't simulate things, like limited play time and other important factors.

P.S. It's almost two weeks passed and we still don't have a patch. Do you think, all this has been done only to pull out attention away from merged queues topic?

[MW Waldorf Statler]

It will not be possible to mathematically calculate the personal ability in a chaotic system with tons of influencing factors, such self-dynamic processes elude any logical calculation for a long time, it is like trying to calculate the direction of a drop of water in the rain and this for everybody Apply rain form

[crazytimes]

Are we doing this or just talking about it until there's no one left playing at all?

[Horseman]

C64 Warrior, on 19 June 2020 - 03:19 PM, said:

Because your personal skill has nothing to do with winning or loosing (there are 23 other people in any given match that effect win/loss) I vote 1A

Statistically, over a sufficiently large sample size the input from the remaining 23 people will average out to become insigificant and the primary remaining factor will be you, your decisions and your actions .

[Xaat Xuun]

Gagis, on 19 June 2020 - 09:53 AM, said:

Why do you think this? It makes no difference for matchmaking, since all players are influenced equally much by luck and other uncontrollable factors, and with time, skill is what makes the difference in if you win more often than not. Any attempt to adjust numbers on a match-by-match basis will lead to less accurate long-term statistics and thus less accurate matchmaking.

just a question,
would bad players being carried up ,would that also be a less accurate matchmaking ?

when I look at 2A, I see no change (very little) to what we have now, but comes with zero-sum, and the good player, on the losing team gets nothing, even though their MS is much higher then the top player on the winning team, they get punished for not being able to carry the rest of the team to a win, they just don't move up, but the losers on the winning team do get moved up.

this is why I like the global options of 1A&B, no one gets carried

now that I have seen Jay Z's proposal 2C, I'm favoring that one, to me it looks like it takes global into account, and does not carry bad players.
I'm just hoping for something that does not carry bad players upward, and keep good players static. I just don't see that with 2 A&B

Edited by Xaat Xuun, 20 June 2020 - 05:05 AM.

[Nearly Dead]

With the current player population, how many skill groups will the matchmaker use in building matches? I suspect you will still see both T1s and T5s most matches as they try to drop matches faster. Maybe in peak hours they might be able to break the players up into two groups. I think three skill groups is probably unrealistic although I also think that would be the minimum needed to improve match quality.

Does anyone think they will have enough players dropping at the same time to divide them into three pools to build teams from? I would be very surprised.

Then add in dropping some matches with a 4 man premade on one team or the other.

I really don't expect match quality to improve much.

[East Indy]

Nearly Dead, on 20 June 2020 - 05:20 AM, said:

With the current player population, how many skill groups will the matchmaker use in building matches? I suspect you will still see both T1s and T5s most matches as they try to drop matches faster.

Players of all skill levels have been dropping together for at least a year, though. Probably a lot longer, but that's as far back as I recall spending my ready-up time checking Jarl's List and spotting low-percentiles almost every match.

From what we've able to see from Paul with the matchmaker "triangle," comparable PSR between teams isn't too difficult to achieve regularly. If PSR can remove major discrepancies between performance and rating, preventing the matchmaker from assigning — for example — two sub-50th percentile players across from two 80th percentile players because it thinks they're all Tier 2, or identifying and balancing for extremely high-percentile players, we have a better system.

This is way, way, way overdue — and Paul has exactly the right attitude now by considering trying a few systems over time. Just because the car won't drive like a Porsche doesn't mean it's free of mainenance!

[Tarl Cabot]

East Indy, on 20 June 2020 - 06:06 AM, said:

Players of all skill levels have been dropping together for at least a year, though. Probably a lot longer, but that's as far back as I recall spending my ready-up time checking Jarl's List and spotting low-percentiles almost every match.

snip

This is way, way, way overdue — and Paul has exactly the right attitude now by considering trying a few systems over time. Just because the car won't drive like a Porsche doesn't mean it's free of maintenance!

For the solo queue, Tier 5 and Tier 1s have not been dropping together, or if they have it has been non-USA primetime. Tier 1 and Tier 2 is made up of both relatively high/high MS players as well as sub 200/190/180 players w/sub 0.85 W/L ratio records. But due to the current PSR values for losses which do not even close to being equalized for the win PSR, said players were able to make their way into Tier 1 and 2 simply through brute force by playing 10s of thousands of games.

If the PSR values had been setup better, as well a breaking up two of the thresholds into three, the disparate would likely not be so extreme. So right now, when the MM seeds and forms up teams for Tier 1-2 players, it is not looking at their avg MS over several seasons/games, only whether or not they are in Tier 1 or Tier 2. MM does not care if one side is made up primarily of sub-200 MS players while the other side is made up of primarily 300+ MS players.

And agree about Paul's current attitude, just wondering where it was years ago.

Edited by Tarl Cabot, 20 June 2020 - 09:06 AM.

[Kosomok]

Jay Z, on 17 June 2020 - 07:00 AM, said:

...but wait....there's more....

Appendix A: Divergence Control Measure

Each and every solution presented will cause the entire playerbase to continually spread out until T3 is less populated and pilots settle towards T1 and T5. This would mean new players, those who play less often and those who change in skill will take a long time to be seen accordingly by the MM after a few months of ANY of these systems. So....let's address that with an easy fix.

Thought of an idea to address the uncontrolled divergence in any in-scope PSR system. What about a regular "Re-scale" of all player PSR ratings on say, a 3 monthly basis to line up with Solaris Reset.

For easy numbers, let's say:
T5: 0-1000 PSR
T4: 1000-2000 PSR
T3: 2000-3000 PSR
T2: 3000-4000 PSR
T1: 4000-5000 PSR

OK so every 3 months scale around middle PSR (2500) the proportion they are away from it (40%).

Maths:

Chosen Variables:
PSRmid = 2500
Scaleback = 0.4

Rescale:
PSRnew = PSRold - Scaleback * (PSRold - PSRmid)
PSRnew = PSRold - 0.4 * (PSRold - 2500)

What does this do? This gives PGI the ability to run a very simple command through the database at the same time as Solaris Reset which will reign in the divergence and pre-seed everyone for the following 3 months. This ensures the reset day match quality is not random where in a full reset match quality would be poor for weeks.

Example:

Pilot "GGpro" is max T1 at PSR 5000. At end of season, calc becomes:
PSRnew = 5000 - 0.4 * (5000 - 2500)
PSRnew = 4000 or borderline T1/T2

Pilot "KillRocks" is bottom T5 at PSR 0. At end of season, calc becomes:
PSRnew = 0 - 0.4 * (0 - 2500)
PSRnew = 1000 or borderline T4/T5

Pilot "BerryMouse" is low T2 at PSR 3179. At end of season, calc becomes:
PSRnew = 3179 - 0.4 * (3179 - 2500)
PSRnew = 2907 or high T3

There you go. This effectively empties T1 and T5 and squeezes the playerbase back to centre in a fair manner within T2-T4 primed for T1 and T5 to refill. Strong pilots can be in T1 immediately separating them from T4 players and below. This should meet PGI's set purpose of PSR in separating new from experienced players.

IOW, regression to the mean.

New players get seeded into T3 by default, non players should slowly regress to the mean. Given the pop sizes, I am not sure that shifting people out of T1 and T5 is, in itself, useful. perhaps if you had a larger player population.