Jump to content

Psr Should Be Changed To Better Reflect Player Skill


71 replies to this topic

#61 Jables McBarty

    Member

  • PipPipPipPipPipPipPipPip
  • 1,035 posts
  • LocationIn the backfield.

Posted 29 January 2016 - 01:29 PM

View Postx MT x, on 28 January 2016 - 06:57 AM, said:

Imho the closest statistics that (we players) can observe that remotely reflects skill is the accuracy stats.

And should be one of the main deal breakers for tier placement.

- Gauss, PPC's, one shot weapons. (Perhaps autocannons too)
- Lasors
- LRM/Streak SRM (Auto aim weapons)


Ever seen former NBA all star and champion Ben Wallace make a free throw? Probably not, because he sucks at them. But as a defensive player it's not his job. A player can have an excellent overall grasp of the game, strategy, and mechanics, but not be a very good shot. So what if they just Lurm from the sidelines? If they are calling out commands, locations, etc., and that's helping the team win, then they deserve to go up. Not every General is sharpshooter.

View PostDuke Nedo, on 28 January 2016 - 05:51 AM, said:

Edit: A parallell (a bit extreme, but just to prove a point) could be to compare to EVE where the FC may sit cloaked on the battlegrid just observing and calling targets and coordinating with other FCs etc, not firing a single shot but clearly being the one largely deciding the outcome of the battle. The skill needed for the "grunts" in Eve is mainly to click on the target and press F1... it's not that bad in MWO, but there is a hint of it, isn't there...? There are many flavors of important contributions that doesn't end up on the score board.


Good analogy. As I see it, the W/L being prime determiner of PSR change encourages killed players to stick in the game and watch their team to help direct them. Also like has been said, scouts, NARCers (who get no points if they don't get the kill while NARC is attached), squirrels, leaders of charges who sacrifice themselves...

View PostMead, on 29 January 2016 - 03:44 AM, said:

One consequence of having a low score on a win be a decrease penalty is that no one will be inclined to capture anything when they could be shooting mechs instead. We'll have three Skirmish modes where the only distinguishing feature is different quantities of bars at the top-middle of the screen.

I sure wouldn't bother with it any more. It's already crappy enough to run around capping and get what is basically a slap in the face for helping land a win.


Yep.

View PostMrMadguy, on 29 January 2016 - 05:03 AM, said:

Nope. Started to lose again. It always happens like that: 2-3 good matches, then lose streak again.

Win a few, then lose a few? I'd say MM and PSR are working as expected.

---

I'm a fan. The only other rating system I have long experience with is Dota 2, which is purely based on the team's W/L. So I like that PSR takes personal performance into consideration, but since it is a team game, it should be heavily based on team performance.

I will say that Dota 2 does give you two separate ratings for Group and Solo queue. So I could get behind that, but my understanding was that Group Queue doesn't really take PSR into account.

#62 Jman5

    Member

  • PipPipPipPipPipPipPipPipPip
  • Littlest Helper
  • Littlest Helper
  • 4,914 posts

Posted 29 January 2016 - 01:42 PM

View PostJernauM, on 27 January 2016 - 09:33 AM, said:

We all understand that PSR is closer to an XP bar than to a true rating of pilot skill. As players generally trend upward towards higher tiers that do not properly reflect their skill level, this flaw in PSR becomes a bigger and bigger problem for matchmaking. Players are being teamed up in matches with a high variance of skill, because the matchmaker is creating those matches with an imperfect understanding of players' "true" skill levels.

If the matchmaker is to depend on PSR, then PSR must be improved. I think the way to improve it is to make it a relative measure, not an absolute score. In other words, if there are going to be 5 tiers, then Tier 1 players should represent the top 20% of the player base, Tier 2 should represent the next 20%, and so on.

To say that a player is skilled is only meaningful if you compare that player's performance to the performance level of other players. PSR should reflect that fact, and so should the matchmaker. There must be some kind of a ranking.


If you want to do it that way you have to do two other things first.

1. You have to introduce a PSR decay for inactivity. This way you don't have higher tiers clogged up by inactive players.

2. You will want to add some sort of bonus pool that accrues over time so people who can't play as often can keep up.

Realistically if you want a meaningful top tier, you would need to do something like Starcraft 2 did with the Grandmaster League. A small pool of 100-200 players who fight over the limited spots. If you made tier 1 just your top 20% the skill range would be enormous.

Edited by Jman5, 29 January 2016 - 01:42 PM.


#63 Pendaelose

    Member

  • PipPipPip
  • Elite Founder
  • Elite Founder
  • 69 posts

Posted 29 January 2016 - 02:53 PM

On my drive home I couldn't stop thinking about how and why we really need to focus on Win/Loss Ratio rather than individual performance and why we need to get away from scoring specific actions for anything other than achievements.

WHY use Win/Loss Ratio Ranking(I will be *glad* to discuss any of these bullet points)
  • Cannot be "farmed" by abusing poorly designed point rewards.
  • Always based on your average performance, not individual lucky/unlucky drops.
  • Does not discriminate against any role.
  • Does not reward any bad behaviors.
  • Always encourages team work over individual scores.
  • Rewards and Punishments are always directly proportional to the impact you have on your team's performance
HOW to implement a Win/Loss Ratio Ranking
  • Only "Current" games should be counted. "Current" meaning the last 30 games, and not more than 2 weeks old.
  • A minimum of 15 games should be "Current" before any rank changes are granted.
  • Win/Loss History is limited in scope to prevent people from coasting among ranks on past performance alone.
  • After each game, if your Win/Loss ratio is greater than 1, add Win/Loss - 1 to player ranking.
  • After each game, if your Win/Loss ratio is less than 1, subtract Loss/Win -1 from player ranking.
  • Converting the current system would simply mean taking your current PSR score and multiplying it by a scale adjustment. Let the system sort itself from there as the great rise and the weak fall.
  • Bracket Tiers by player percentile. The highest and lowest tiers should be very narrow with most people falling in the middle tiers.

The real meat to the idea is that "If you are on winning teams more than losing teams, you must be doing something right".

What a person is doing "right" is almost impossible to quantify. Maybe you're a hell of a good shot with a solid build. Captain Alpha cores again! Maybe you're an amazing scout, or a player who watches the map and makes good calls over the VOIP. Maybe you're amazing at spreading that damage, a real bullet sponge and always where you need to be. Maybe you're a squirrel. Maybe you run ECM and escort your Assault class buddies while shooing away the squirrels. Hell, maybe your only skill is knowing who to group with... It doesn't matter... the truth is, if you have a Win Loss rating greater than 1.0 you help your team win more often than not.

I don't care what a player's Kill/Death Ratio is. If you get 5 kills every match but somehow cost your team then win 80% of the time I don't want you on my team. I'd rather have that great scout who knows when to pop a UAV and when to call targets. I've got a 2.27 K/D ratio, I'll gladly carry his weight in damage if he can tell me where I need to be.

#64 pyrocomp

    Member

  • PipPipPipPipPipPipPipPip
  • Philanthropist
  • 1,036 posts

Posted 29 January 2016 - 03:30 PM

View PostPendaelose, on 29 January 2016 - 02:53 PM, said:

  • Only "Current" games should be counted. "Current" meaning the last 30 games, and not more than 2 weeks old.
  • A minimum of 15 games should be "Current" before any rank changes are granted.
  • Win/Loss History is limited in scope to prevent people from coasting among ranks on past performance alone.
  • After each game, if your Win/Loss ratio is greater than 1, add Win/Loss - 1 to player ranking.
  • After each game, if your Win/Loss ratio is less than 1, subtract Loss/Win -1 from player ranking.

When a player will buy a new mech he will almost sertainly drop in performace for the duration of leveling a new mech. After that if the new mech was in the weight the player is familiar with he for minimum of 30 games (depends on how deep a hole he dug) will be seal-clubbing less fortunate players. Not good at all as on many occasions there newly-bought mech on the battlefield. Your system might be good in the situations where there is no unlockable gear upgrades, but, I'm afraid, not here.

#65 Pendaelose

    Member

  • PipPipPip
  • Elite Founder
  • Elite Founder
  • 69 posts

Posted 29 January 2016 - 04:28 PM

View Postpyrocomp, on 29 January 2016 - 03:30 PM, said:

When a player will buy a new mech he will almost sertainly drop in performace for the duration of leveling a new mech. After that if the new mech was in the weight the player is familiar with he for minimum of 30 games (depends on how deep a hole he dug) will be seal-clubbing less fortunate players. Not good at all as on many occasions there newly-bought mech on the battlefield. Your system might be good in the situations where there is no unlockable gear upgrades, but, I'm afraid, not here.


None of those things change in any other ranking system. It takes several games to change your rank in every system. The difference here would be that if you are only week for a couple days your ranking would only dip a little instead of a lot. A ranking system where ranks change faster would mean he spends drops even further and then clubs harder until he's back up. This isn't a problem with my system, but rather an intrinsic part of slow or fast ranking changes.

#66 Tier 1 Smurf

    Member

  • PipPipPip
  • Death Star
  • 61 posts

Posted 29 January 2016 - 04:47 PM

All PSR really does is keep the underhive separate from the T1 golden master race, and allows real T1 players to find matches faster (by filling the matches out with psuedo T1 players).

#67 MrMadguy

    Member

  • PipPipPipPipPipPipPipPipPip
  • 2,308 posts

Posted 29 January 2016 - 11:04 PM

View PostMrJeffers, on 29 January 2016 - 10:21 AM, said:



PSR is nothing like elo or true skill and performs nothing like them. You are so blindly wrong it's not even funny. There is no prediction or relative performance to other players involved.
Your rank can't drop on a win, even if you are AFK the entire match. The only time that PSR drops is on a loss, but if you perform well enough on a loss you can get a slight increase. This is shown in Paul's graphic that has already been posted in this thread.

There are *fixed* hardcoded match score levels that determine the outcome. On a Loss a match score of less than 250 is a decrease. 251-400 is a neutral performance, and 401 is an increase.
On a win greater than 100 points is an increase but only Paul/PGI know the tiers of increase, but one could assume they are the same scale as the losses.

It in no way is performing any measure against other players like elo systems, it is completely unrelated and why it's referred to as a glorified XP bar because it's a fixed table of match score vs win/loss determines your point gain/loss in tier. That is the only criteria.

It would be better for it to be TrueSkill. At least is has more sense. I have analyzed current PSR system. I looked for spot, at which you may stabilize - average Match Score, you need to maintain, while your W/L = 1. And it seems, that you will advance, if your Match Score > 250 and will drop if it < 100. 100-250 interval in W/L dependent. So that means, that lower skill players, for whom Tier 1 is way to high, are doomed to sit at 100 Match Score level forever??? That's what actually happens to me - I sit at 100 Match Score level and can't move anywhere. As soon as I reach MS = 100 level - I stop dropping.

You should understand, that 100-250 MS cap - is way too low. Actually system is designed to prevent good balanced games, where players live long enough to earn enough MS, from happening. As long as you start getting better matches - you start advancing and getting terrible matches again. Only streaks of stomps. Again, again and again.

Edited by MrMadguy, 30 January 2016 - 12:08 AM.


#68 MrJeffers

    Member

  • PipPipPipPipPipPipPip
  • 796 posts
  • LocationThe Rock

Posted 30 January 2016 - 08:45 AM

View PostMrMadguy, on 29 January 2016 - 11:04 PM, said:

It would be better for it to be TrueSkill. At least is has more sense. I have analyzed current PSR system. I looked for spot, at which you may stabilize - average Match Score, you need to maintain, while your W/L = 1. And it seems, that you will advance, if your Match Score > 250 and will drop if it < 100. 100-250 interval in W/L dependent. So that means, that lower skill players, for whom Tier 1 is way to high, are doomed to sit at 100 Match Score level forever??? That's what actually happens to me - I sit at 100 Match Score level and can't move anywhere. As soon as I reach MS = 100 level - I stop dropping.

You should understand, that 100-250 MS cap - is way too low. Actually system is designed to prevent good balanced games, where players live long enough to earn enough MS, from happening. As long as you start getting better matches - you start advancing and getting terrible matches again. Only streaks of stomps. Again, again and again.


And you need to understand that no match maker can prevent stomps in MWO because of the lack of resapwns and limited team health. Terrible matches are 100% because of the lack of teamwork and the only way a match maker could fix that would be boosting/healing a losing team. Something that you just can't seem to understand or simply refuse to accept. Someone else came up with a good example that hopefully is clear enough for you to understand:

View PostDaZur, on 29 January 2016 - 01:38 PM, said:

Let me make it simple for you MM theorists...

Think of a match as a game of tug-of-war. Each side has 12 participants. Regardless of the skill level of the participants so long as there is 12 on 12 the match is a essentially equal. When one side loses a player the other side gains it's contribution strength. Typically, so long as the differential from one side to the other doesn't exceed roughly a 1/4 differential, each side has an equal opportunity to win.

It's when one side gains in advantage greater than 1/4th of the total participants the combined force strength overwhelms the now deficient side. This is what is called "combat force attrition".

This is true whether it be tier 1 or tier 5. The only difference is the higher the tier level, the quicker the team with the numerical superiority takes advantage in the disparity of force strength.

Mathematically, this is a predictable outcome based on numerical advantage. In short there is nothing a matchmaker can do to mitigate a role once it is in motion.



Elo was a much better system for actually rating _skill_ because it actually compares your performance to your opponents. PSR is not a skill ranking, it's an experience ranking and really should have been called PER.
A key example of the difference - Top tier players in elo system plays bottom tier players and wins - players elo remain relatively unchanged. Bottom tier player wins and top tier players lose and the top tier players see a huge drop while the bottom tier sees a corresponding huge gain.
In PSR Top tier players play bottom tier players and win - Top tier players see a huge increase in rating because of inflated match score. Bottom tier players see huge decrease because of the poor match score. Top tier players play bottom tier players and lose - top tier players see no change or slight gain because of high match score, bottom tier players see slight gain or no movement because of lower match scores.

Edited by MrJeffers, 30 January 2016 - 11:12 AM.


#69 wanderer

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 11,152 posts
  • LocationStomping around in a giant robot, of course.

Posted 30 January 2016 - 11:35 AM

View PostTier 1 Smurf, on 29 January 2016 - 04:47 PM, said:

All PSR really does is keep the underhive separate from the T1 golden master race, and allows real T1 players to find matches faster (by filling the matches out with psuedo T1 players).


It really doesn't even do that, because the tier bars are virtually impossible to not advance up.

This means the odds of people ending up in T3 and becoming T1 fodder is near 100%, regardless of actual skill. It only requires enough time played.

The minimum match score for getting a PSR gain need to go up as tier rank does, the minimum to avoid a PSR loss needs to go up along with it, and the effects of a really bad match score need to have larger effects as your tier rank does. A T5 PSR-down score from a T1 player should hurt far more than a barely-missed-it effort at the same rank.

#70 Tier 1 Smurf

    Member

  • PipPipPip
  • Death Star
  • 61 posts

Posted 30 January 2016 - 12:05 PM

View Postwanderer, on 30 January 2016 - 11:35 AM, said:

It really doesn't even do that, because the tier bars are virtually impossible to not advance up.

This means the odds of people ending up in T3 and becoming T1 fodder is near 100%, regardless of actual skill. It only requires enough time played.

The minimum match score for getting a PSR gain need to go up as tier rank does, the minimum to avoid a PSR loss needs to go up along with it, and the effects of a really bad match score need to have larger effects as your tier rank does. A T5 PSR-down score from a T1 player should hurt far more than a barely-missed-it effort at the same rank.


It's like the system can't lose.

If T1 was full of noobs because PSR is bad, then all the noobs moving up PSR will be right at home and MM is working. If the player is bad and yet still moves up to T1, and it's full of leet players where the baddie gets pwnt, then the system will move them down.

#71 wanderer

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 11,152 posts
  • LocationStomping around in a giant robot, of course.

Posted 30 January 2016 - 01:12 PM

View PostTier 1 Smurf, on 30 January 2016 - 12:05 PM, said:


It's like the system can't lose.

If T1 was full of noobs because PSR is bad, then all the noobs moving up PSR will be right at home and MM is working. If the player is bad and yet still moves up to T1, and it's full of leet players where the baddie gets pwnt, then the system will move them down.


I don't really see anyone who actually sits there saying "I KEEP RANKING DOWN!"

Because statistically, it's virtually impossible. Even with people getting dozen+ losing streaks, the trend is "gain". Given that, the general result is going to be T2-T1 will end up with people who long ago plateaued at a given skill level, but end up carried enough to more than compensate for bad performance. I mean, I already end up in games with T1s if I pick a low enough % weight class on quick play...and there are plenty of T3's already have literally skidded face-first into that.

They're still marching up the tier bars. If your performance is enough to get you up from Rank 5, eventually it'll be enough to get you to rank 1 in the current system. The only difference will be time spent getting there, but the march is inevitable. An actual PSR system should require effort just to hold your spot, exceptional success causing it to rise, and failure to do so causing noticeable slippage.

It doesn't. A system that eventually puts bads and goods in the same "top" tier is just a delaying tactic for the pre-PSR matchmaking to return.

#72 Scharfschiesser

    Member

  • Pip
  • 18 posts

Posted 04 February 2016 - 06:02 AM

Just rename it from "PSR" to "RNG", because that is what it is... a random number generator

It generates a number based on your win/loss ratio and as this game is 12:12 the "influence on outcome" for every single player is 1/24 - or in other words: random!





1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users