357 replies to this topic

[General Solo]

50 50, on 21 June 2020 - 07:13 PM, said:

I would expect a high performing player to do well regardless and be getting good scores win or lose.
In a group, that win to loss ration might improve, but not necessairly their match score.
A good player grouped with other good players may see their score decrease in comparison as they are no longer the only one playing at that level and part of what made up their match score is now being taken by those players.

As you point out:

Precisely, its why match score is better than w/l as it takes out the group skill from the player skill rating.

50 50, on 21 June 2020 - 07:13 PM, said:

.....
Bad matches happen however and looking at the numbers put forward by JayZ and the other contributers there, it is based on individual performance with a slight weighting for win/loss.
So if you are a good player and put up a good score, you would still see that increase in PSR on a loss and a player getting a bad score on the winning team would still get a decrease.

Straight up, I do not like the PSR changes based solely on your team winning or losing. ie. The losing players have no hope of gaining PSR.
The 1A example seemed a lot fairer as it was purely based on individual performance but you might argue that if players did so well but still lost, why should they get the same level of increase?

However, that's why the proposal in the external spreadsheet seems fair.
It is taking into account the individual player performance.
It gives a slight bonus for the win.
It also seems to cater for a stomp scenario with the way it is working it out.
But if you have a lousy game even on the winning team and even in a group, you will see a reduction in PSR.
Likewise it is still allowing for a player to have a blinder of a match on a losing team and get the increase.

A follow up question might be how are groups going to be placed in the match maker?
Is it going to use the player's tier and average over number of players in a group?
Would it use the accumulated PSR total and then averaged over number of players in a group?
Will being in a group add some weighting to that?

I see 1A doing those things as well, do 500MS win or lose that player did 500MS of skill and should go up PSR accordingly and vica versa.

I'm not bagging JayZ's model, never have, she's hawt.
Just like I said that I see that 1A will do the trick, its logic is simple and fair which makes the maths simple as well.
Top MS players go up PSR win or loss and Bottom MS players go down PSR win or loss, until they find their equilibrium and peers.

I still remember when the current MM was first released years back. It worked very well for the first year or so. My unit mates never met EMP unless I was in the group. True story.

So even 1A which has even less bias than the current system, should be good enough for about a year and a half.
Just reset it then and make an event.
MM reset Season 3.

[C64 Warrior]

Horseman, on 20 June 2020 - 03:07 AM, said:

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 .

and this is exactly why we need a system that measures OUR individual performance and not just win rate/our luck at dropping on the winning team. Our PSR should be based on our actions not the sum result of how the other 23 people performed.

[Gagis]

C64 Warrior, on 21 June 2020 - 11:23 PM, said:

and this is exactly why we need a system that measures OUR individual performance and not just win rate/our luck at dropping on the winning team. Our PSR should be based on our actions not the sum result of how the other 23 people performed.

The random elements are equal for everyone. Luck is blind. Random is random. Over a large sample of matches played, W/L is the most accurate possible way to measure your success. Match Score based corrections to make people feel good will be just that, adjustments for the sake of making people feel good about a single individual match, but they will unavoidably reduce matchmaker accuracy as well, since match score can at best roughly approximate the ground truth that is your actual chance of winning or losing.

What kind of schools have people gone to, for so many people in this thread to have absolutely no understanding of how chance and statistics work? Pretty much everyone who has argued in this thread against W/L has done so on fundamentally flawed premises. Just do the damn math.

I'm sorry for being so snappy at this point, but it is getting seriously frustrating to see same false statements repeated page after page after page. Your gut feeling about what seems right on the results screen of a single game are not logic. They are just your gut feeling.

[Adrian0815]

I don't understand why only a small part of the matches should be used to determine the personal performance of a pilot and to raise or lower him to the appropriate tier level.
A good pilot in an average team will seldom decide alone about victory or defeat. The same applies to the below average pilot. To determine the rating, every match must be taken into account and only then can statistically sound results be achieved.

In the systems 2a and 2b only good players would be moved up in case of wins and worse players would be moved down in case of defeats. This means that a correct rating can only be achieved after many, many matches. In the meantime, the matchmaker often assembles unequal teams. It would remain with a high stomprate, which will surely frustrate some. Also the mood in the matches would become even more dogged, "noobs" would be put down, because they prevent a legitimate rising of overpervormers.

Therefore I see 1a as the only viable system for the future.

[Nearly Dead]

Tomorrow is Tuesday. Do it. Anything. I want to go back to playing or cut my losses and I really don't care much which it is anymore.

[RRAMIREZ]

C64 Warrior, on 21 June 2020 - 11:23 PM, said:

and this is exactly why we need a system that measures OUR individual performance and not just win rate/our luck at dropping on the winning team.
Our PSR should be based on our actions not the sum result of how the other 23 people performed.

Point being: what is "individual performance" in a 12v12 game.


Bringing victory through strat, coms, calls, decision making being not really measured in MS, that part of the "individual performance" is badly (or not at all) taken into account.
It is in WLR

For instance, I, average+ potatoe, have a correct/good aim, torso twist, situationnal awarness.
But I play the game as a casual, team follower... my decision making is not that great, my coms are basics, and my english is not fluent (and I tend to be too much aggressive)
What is my individual performance, globally?
Difficult to say but I have the feeling that if I'm ranked by average MS only (core 1a), I will be overrated.
[color=#222222]There's still problems with pure WLR though, low matches numbers being one.[/color]

leaderboard:
Season 47 (116 games) - AMS 265 - 3178 position - WLR 0.93 - 7547 position
Season 46 (278 games) - AMS 253 - 4742 position - WLR 0.89 - 10600 position

my 2 cts on individual performance.

ps:
I'm high T2 and play Medium/heavy mainly

Edited by RRAMIREZ, 22 June 2020 - 05:07 AM.

[ERSmurf]

Nearly Dead, on 22 June 2020 - 04:24 AM, said:

Tomorrow is Tuesday. Do it. Anything. I want to go back to playing or cut my losses and I really don't care much which it is anymore.

@Nearly Dead: There is an ingame maintenance alert for tomorrow evening my local time. I do not know what they will do, but they are doing something. I guess we stay tuned here on the forum for more info on tomorrow...

[Xaat Xuun]

Bowelhacker, on 21 June 2020 - 04:42 PM, said:

Apologies if this has been mentioned in one of the previous 12 pages I don't have time to read through, but is this being applied to faction play results as well?

I'm pretty sure, this has nothing to do with Faction play. Faction is a different , there, every one should be grouped before the match, filling empty slots with random solo/small group players unable to find and join a group, and what their skills are just a random chance they know what to do,

different world between Quick play and Faction Play

Quick play is more Solo dropping in a team (used to have a group drops, but now just 2-4 player group)

I'm sure someone can better explain then I did

Edited by Xaat Xuun, 22 June 2020 - 11:29 AM.

[Adrian0815]

RRAMIREZ, on 22 June 2020 - 05:03 AM, said:

Point being: what is "individual performance" in a 12v12 game.

Bringing victory through strat, coms, calls, decision making being not really measured in MS, that part of the "individual performance" is badly (or not at all) taken into account.
It is in WLR

Just like, for example, an 11vs11 game - football (or soccer as they call it in America)
There values are recorded for each player, such as mileage, duel strength, pass quota, goal finishes, goal preparation, fouls, etc. In this way, the good players in the losing team are also identified and then bought up by teams with purchasing power in the transfer periods.

In MWO, player actions are also recorded and evaluated (e.g. kill, damage, scouting, capture, flanking, lance in formation, etc.). Certainly things like good communication and general strategy help to win, but victory or defeat is recorded in the match score anyway.
Of course, one can think about whether the current evaluation of the actions is justified (e.g. damage is "too much value") and what might still have to be recorded to identify good pilots.

[Nightbird]

OZHomerOZ, on 21 June 2020 - 08:53 PM, said:

That math is but a tool to do a job, applied to a problem
If the problem is not understood guid the maths won't fix it.

Rate Pilot Skill (all situations - Grouped, Ungrouped, any mech/map/mode/build ) using one number to rate the pilot skills in all conditions so MM can do its thing.

Look at the first post in the last thread. The problem is reducing stomps. Not one solution has a mathematical proof showing stomps going down.

Forget 1+1=5, we're dealing with 1+1=Ohio level of thinking here.

[Ridir Semii]

I had to stop reading.... there are so many convoluted posts that have nothing to do with which version you would like and potential match score kicker change suggestions...… I will try to stay on topic:

I personally prefer 1A with the larger neutral zone (non-changing of psr)

As for any MS kickers, I would say that since w/l is pointless to be used in the actual MM, but IS very important to the overall equation, change the win/loss kickers by 50-75% respectively, meaning:

a win should carry 50-75% more ms increase than it currently does, and vise versa for losses.....

[Cluster Fox]

50 50, on 21 June 2020 - 07:13 PM, said:

... A good player grouped with other good players may see their score decrease in comparison as they are no longer the only one playing at that level and part of what made up their match score is now being taken by those players.

Exactly what 50 50 said. That happens to me for instance. When I drop solo I get very high match scores. I try to team play as much as I can but do my own thing when it doesn't make any sense. Dropping in group (we are non-elite players), we coordinate better and share the workload better, usually trying to get the pugs to buy into our strat. Our matchscore is lower but we win more.

TL;DR;
Non-elite group player here:
Solo - Our match score is higher, W/L ratio is lower.
Group - Our match score is lower, but our W/L ratio is higher.

So all of those arguing for 1A on account it will separate the groups from the solos. Sorry guys, it'll have the opposite effect except for a few elite groups.

Keep in mind with all of those systems, there is no equilibrium. The more matches played, the more it diverges from the middle.

As Nightbird pointed out, none of these has proof of work on expected results for reducing stomps.

Edited by Cluster Fox, 22 June 2020 - 07:57 AM.

[Anomalocaris]

Gagis, on 22 June 2020 - 01:07 AM, said:

The random elements are equal for everyone.

This is not true in MWO. It is less true now with self-grouping 2-4 man drops in game. Even if it were true, the number of matches required to get proper ranking is greater than the average number of matches played each season by individual players.

As I've said before, the idea of reworking PSR was admirable, but with PGI's limitations as expressed, it's just not going to really matter at this point.

[Cluster Fox]

So, if Core 1A/B ends up implemented. People should be ready for either:

A. More stomps.
B. A -very- large score spread between Win/Loss. Possibly negative matchscores on bad losses, low scores on all losses.

Let me explain:
To factor anything close to WLR out of a match's outcome, you can only have a sum of 0 for a WLR of 1.0.
Then the deviation becomes (WLR - 1) * matches played. Let's call this pseudo WLR. Since we cannot get WLR, pseudo WLR is the best approximation. I'll refer to that desired pseudo WLR as (W+L)n

To account for (W+L)n, Core 1 needs to have a delta of Loss Bonus and Win Bonus that is significant vs the match score.

So the important value for Core 1 is the spread between Win and Loss bonuses.
Here's an example for 4 player. 0.9 to 2.0 WLR. WLR, 2.0 representing a group player.

Proof (using Core 1 logic):
If win = +100MS, loss = -100MS, Delta is 200MS, match avg 0MS (example kickers)
WLR 1.0 : (W+L)n = 0 * matches played = 0
WLR ~0.9 : (W+L)n ~= -10 * matches played => player goes lower in PSR until he starts winning more, trending towards 1.0 WLR. Worth about 20 DMG per match.
WLR ~1.1 : (W+L)n ~= +10 * matches played => player goes higher in PSR until he starts losing more, trending towards 1.0 WLR. Worth about 20 DMG per match.
WLR ~2.0 : (W=L)n ~= +100 * matches played. Worth about 200 DMG per match. Ok, not bad, we get some upward movement here but a good pug with 1.0 WLR will move with them.

If win = +50MS, loss = 0MS, Delta is 50MS, match avg +25MS (current kickers)
WLR 1.0 : (W+L)n - 25 = 0 * matches played => Player doesn't move.
WLR ~0.9 : (W+L)n - 25 ~= -2.5MS * matches played => Or worth about 5 DMG per match.
WLR ~1.1 : (W+L)n + 25 ~= +2.5MS * matches played => Or worth about 5 DMG per match.
WLR ~2.0 : (W+L)n + 25 ~= +25MS * matches played => Or worth about 50 DMG per match.

This brings the question of: How important is WLR? Well, more so than match score according to Ref A. 200MS spread is significant, yet still has a limited effect. We're talking about 10% of an average player's match with 200MS spread. If the WLR is 0.99 the (W+L)n effect is statistically very small and therefore pretty much hidden in the PSR global score. With a 50MS spread, it would take a 0.99 WLR a 1000 matches for his WLR shift to be worth 250 match score !!!

Again, when I play grouped up for instance, my damage is on average 100-150 lower than solo because we share the high damage output among the group better. With that into account, a 50MS spread between Win and Loss means I actually go relatively down in PSR when grouping up with Core 1! Even if we win a -lot- more.

----

That being said, Core 2A, 2B and JayZ's A,B suggestions take Win/Loss into account on the PSR shift directly and are based of the team's performance, no problem here for (W+L)n accounting from matchscore. This is preferable since (W+L)n becomes independent of matchscore kickers for win/loss. With Core 2's JayZ included, WLR vs avgMS gains can be adjusted separately, without playing with the kicker values for 3 months. And you can deal with large PSR shifts while still making players feel good and fairly assessed at the end of the match.

However, JayZ's 2C is a hybrid solution. Slightly less favorable since PSR formula and Matchscore are not independent anymore.

Ref A. Nightbird modelled matches over time and PSR, findings were that WLR is the #1 parameter to reduce stomps. So it should be weighted heavily because heavy shifts from (W+L)n will mean a discriminating PSR (good discrimination here) and better balanced matches quicker.

Edited by Cluster Fox, 22 June 2020 - 10:27 AM.

[Brauer]

2C or 2B.

1B would also be a decent option if getting a W netted a substantial enough match score boost.

[Capt Deadpool]

I suspect a number of players, whether they realize it or not, are inclined towards choosing the system they believe will allow them the best odds of retaining the shiny 'Tier 1' badge, even if that system results in more stomps and a worse playing experience for them.

[L1f3H4ck]

Capt Deadpool, on 22 June 2020 - 10:09 AM, said:

I suspect a number of players, whether they realize it or not, are inclined towards choosing the system they believe will allow them the best odds of retaining the shiny 'Tier 1' badge, even if that system results in more stomps and a worse playing experience for them.

Yeah, that's the way. Alleging ulterior motives is how you keep a conversation constructive.

[Capt Deadpool]

Dakkalistic, on 22 June 2020 - 10:25 AM, said:

Yeah, that's the way. Alleging ulterior motives is how you keep a conversation constructive.

Well, I don't think it is necessarily a conscious decision, though in some it might be. Commenting on biases determined by basic human psychology is not unconstructive. And I don't think you would be able to provide any evidence that 100% of the players choosing a system here are definitively more concerned with competitive matches than achieving an e-peen tier.

When you combine this with the already stated fact that it seems some people don't have a grasp of basic statistics, it would not be prudent to exclude this information from the decision-making process.

Edited by Capt Deadpool, 22 June 2020 - 10:53 AM.

[Vindicated]

OZHomerOZ, on 21 June 2020 - 09:36 PM, said:

Precisely, its why match score is better than w/l as it takes out the group skill from the player skill rating.




I see 1A doing those things as well, do 500MS win or lose that player did 500MS of skill and should go up PSR accordingly and vica versa.

I'm not bagging JayZ's model, never have, she's hawt.
Just like I said that I see that 1A will do the trick, its logic is simple and fair which makes the maths simple as well.
Top MS players go up PSR win or loss and Bottom MS players go down PSR win or loss, until they find their equilibrium and peers.

You might want to explain why you consider match score to be a fair (unbiased). It obviously favors certain type of mechs.

• TLDR will say first, there is a strong bias towards backline mechs. As a backline mech (such as 6 AC2), I tend to be one of the last mechs alive and I can pick up higher damage and more kills (broken mechs). With 6 2's, my purpose is to not stop firing, which adds up over the time span of typical games if my team does not die immediately (they are just as important).
  
  • Key point: Regardless of MS recalculation, if KILLS and components destroyed are being factored in, I still come out on top. If KMDD is being factored in, I still come out on top since I will have done the most damage.
• One thing everyone brings up is LRMs. I have bad (0.5) W/L in my LRM boat and I rarely end up with anything less than a equal sign, even when I did next to nothing (~200-300 damage).
  
• I have above average W/L (1.5) in ATM Vapor Eagle 3. Even so, I only get 400 MS or so on wins because I am focused (rightfully so) after being discovered, ideally after taking down or at least crippling one big mech or more.

Please explain my current season stats any other way. I play more than enough games to say I am average, yet I have higher MS than players with better stats and will win if we have 12 of me and 12 of them. Since it's recent enough, I can tell you my mechs of choice this season which break 500 MS regularly (can even hit 700) on wins have been 1. 6 (U)AC2 Rifleman IIC 2 or 6AC2 + AMS (not for MS farming, it's because I need unusually high face time) Mauler MX-90, and 2. 6LL + AMS Stalker 4N. All back line mechs.

Rank Pilot Name Total Wins Total Losses W/L Ratio Total Kills Total Deaths K/D Ratio Games Played Average Match Score
41 Vindicated 318 310 1.03 566 397 1.43 629 312

Between my 6 2's explanation for my high MS and your suggestion that team coordination is irrelevant, go enjoy your 12 man Div1 Solaris (significant amount of 6 2's) tournament. Go to different grid squares, finish opponents off individually. Then winner takes leftovers.

Guys don't hold a firing line, the top tier players don't do that because they won't get as many kills. Oh and don't stand in the same grid as me, that's going to bring down my MS and make me look like I suck. Unless you really hate someone, then we'll go focus fire them and they'll be dead in 3 or less seconds (we've done the math and tested this) and they'll look like they suck.

Edited by Vindicated, 22 June 2020 - 10:52 AM.

[Bud Crue]

Capt Deadpool, on 22 June 2020 - 10:09 AM, said:

I suspect a number of players, whether they realize it or not, are inclined towards choosing the system they believe will allow them the best odds of retaining the shiny 'Tier 1' badge, even if that system results in more stomps and a worse playing experience for them.

Really? I just don't get that. Presumably we all want the best quality matches we can get (i.e. less frequent stomps/rolls). We can all look at Jarl's. As such, we all know (or ought to know) who the real T1s are vs the rest of us. All of us who are currently T1 but not remotely top players, ought to be longing for placement into the appropriate Tier, regardless of the label. I'll take an Under hive badge if that helps get me paired up against the people I really ought to be playing against/with (though honestly I think none of this will matter, as I expect the population is so low that the gates will still have to be essentially wide open now that the MM has groups to contend with).