Psr Update And Hold On Patch.
#361
Posted 10 June 2020 - 09:47 PM
For those saying that, since winning the match is the point of the game, this makes a lot of sense. So too for those that had issues with match score based on the fact that narc, scouting, baiting, etc. don't significantly contribute to match score but can have a large impact on winning.
For those (myself among them) who have issues with losing PSR on a loss where you personally performed reasonably well, remember that one match won't have a significant effect. We're talking about calculating an average over multiple matches, and that it's not static. If you start going down in tier you'll hit a point where your personal contribution makes a significant difference in team win/loss, and start rising again. By the same token, a player getting carried a lot will rise to a point where EVERY player's contribution impacts win/loss, and start falling. Look at Nightbird's data, it clearly shows that, over 10,000 matches, the W/L ratio for every pilot begins to approach 1.0, meaning that the pilots end up in the proper tier.
Just my $0.02.
#362
Posted 10 June 2020 - 09:51 PM
To the creators, thanks for acknowledging that something in match maker needs to change.
It is fairly obvious that the majority of players realise that the current PSR is a "Player Experience" rating not a player skill rating as intended. With that I do think that this current suggestion is a step in the right direction, but are you going far enough? I'm sorry but I can only describe this new suggestion as a "Player Average Match Score" system, it is still not a player skill rating. In my opinion, this game has far to many variables and you are fighting a loosing battle trying to compress all those variables into just one statistic. (again a step in the right direction)
Under this new suggestion, scenario, 2 players with same PSR, one is better at assaults, other is better at lights, if they both choose a light mech for a game, the match maker will have them as even when in reality they wont be and the team with the light specialist will have the advantage.
Have you considered Average Match Score "Per Mech"? That is were I would start at a minimum and if that doesn't achieve an improvement in balancing then go further with match score per map and match score per game mode.
Again, thanks for your efforts and considerations in trying to reduce the "stomp" rate.
#363
Posted 10 June 2020 - 10:42 PM
Surn, on 10 June 2020 - 09:12 AM, said:
yes, I wanted to show a decent game and an extreme bad game, I record most of my matches, so I could put up an extremely good game, or a slightly bad game also.
Mostly, I want to fine tune the numbers
So in that chart with RPI values all teamplay actions and going for objectives are somehow low? Like player scores very low for teamplay actions? I'm not sure if I fully understand that.
#364
Posted 10 June 2020 - 11:51 PM
Mordenthral, on 08 June 2020 - 04:41 PM, said:
Otherwise, we should just get rid of the other game modes and make all matches Skirmish. Which wouldn't be as much fun.
captureassist
capture - Gain for capturing a capture zone.
capturepulse - Gain for time you are capturing in a capture zone.
firstcapture - Gain for capturing the first capture zone in a match.
powercell_pickup - Gain for picking up a power cell.
powercell_dropoff - Gain for dropping off a power cell.
kill_powercell_carrier - Gain for killing a power cell carrier.
I do think we need at least the relative weighting for the different triggers.
To whit, the scores for "this match type" should be boosted relative to "kill everything" scores if that's not the match type?
That makes sense to me.
If the point of a match is win/lose rather than "get kills", it also makes sense that "Win" should be substantive relatively speaking too. But too much and it makes "kill everything" more viable again.
PGI: Is it possible to get the relative trigger weightings, and the counts for the individual triggers for all 24 players for say 2-3,00 matches, and a "group" indicator, match duration, Live/Die, and the "win trigger" (killed all, got points, etc)?
e.g.
Match 1,Player 1, Solo, 1,2,3,4,5,6,7 (trigger scores)
Match 1, Player 2, G1, 2,3,4,5,6,7,...
The "player" ID should be obfuscated, but consistent between matches.
"Group" can be fresh each match.
Blakemiles, on 08 June 2020 - 04:48 PM, said:
This assault lurmer dislikes dying first by virtue of being too slow to keep up while everyone else charges off.
#365
Posted 11 June 2020 - 12:11 AM
Akillius II, on 08 June 2020 - 05:24 PM, said:
1. Anyone on either team looses the most PSR if they are perma-AFK, -OR- under 50 match score.
However PGI's servers must detect and take into account when the servers hiccup and drop as many as 22 players per match.
2. Anyone damaging team mate within first 30 seconds of match starting would instantly loose -3 PSR at match end irregardless of "performance".
And -10 PSR if they "accidentally" kill the (light) team mate within first 30 seconds of match starting.
3. Killing a friendly at any other point during the match gives an additional -2 PSR.
(yes accidents happen, but if your half as good as you think then no worries, you'll +1000 matchscore and +5 PSR anyways)
#2 - EXCEPT if it's collision damage.
Every d**n match I have in an assault, I start facing the wrong way, and while I'm turning around someone almost always crunches into me.
#3 - -(5 + (5 * (team kills)^2)% of (match PSR share / 24), just in case someone goes "killed one? Might as well kill them ALLLLL!!!!!!"
#366
Posted 11 June 2020 - 12:17 AM
Gagis, on 10 June 2020 - 03:43 PM, said:
Using a ratio of your winning performances and losing performances prevents the PSR from exploding away from the mean and instead stabilizes over time towards a number that properly describes your performance regardless of if you have played a thousand matches or a million matches.
Interesting point, but perhaps flawed? Consider this - isn't the WLR the *outcome* of the MM? Ideally the WLR will trend to around 1.0 for most people who are in the correct tier: they win half the time, they lose half the time because the other team is of similar skill.
About concerns of using MS for PSR, and the PSR becoming 'divergent' with the zero-sum proposed: If it happens, this would only affect the extreme ends of "below Tier 5" and "above Tier 1" players. Exceptional players like these will likely always exist. But note also, that if these exceptional players are matched against similarly exceptional players the gameplay would still be balanced - maybe these people can form the extreme top and bottom tiers. This leads to the other factor of what ranges of PSR form each Tier, which is another variable that can be adjusted.
(Edited for clarity.)
Edited by RCore, 11 June 2020 - 12:20 AM.
#367
Posted 11 June 2020 - 12:50 AM
1) Do we want PSR to be W/L dependent?
Yes - Make values different for win and loss.
No - Make them the same.
2) Do we need true zero-sum or not?
Yes - change PSR according to relative performance vs other players.
No - change PSR according to relative performance vs fixed value.
3) How fast do we want player to climb/drop?
Set PSR change values accordingly.
Everything else - is purely about MS calculation. What actions mean more skill and what don't. At the end precise MS calculation isn't that important now. Except may be reducing DMG component. Because it can never be perfect anyway.
So we just need to answer to this 3 questions above and make changes as soon, as possible.
Another thing, that would be nice - per 'Mech PSR modifier. But such system is way too hard to implement.
Edited by MrMadguy, 11 June 2020 - 12:51 AM.
#368
Posted 11 June 2020 - 01:02 AM
I'm too damn old to be reading all of this when I should have been out there melting armor instead.
My thoughts on the PSR/MS update/calculations:
1) Paul, please give us the exact Match Score(MS) formula used associated with all the triggers you had listed.. Knowing all the triggers is fine and dandy, but not knowing exactly how they're currently being used to come up with the final MS number is no help to people trying to tweak the formula... Thanks in advance
2) I personally like the one suggestion where you're ranked 1-24 by MS (i.e. you're ranked on how you did versus the other 23 people playing in the current match) ... Top 8 - PSR goes up by some factor, Middle 8 - no change to PSR, Bottom 8 - PSR goes down by some factor ...
3) Take W/L out of the MS calculation ... it depends too much on whether you win the pug lottery or not ... If you REALLY have to have W/L in the calculation, then make it some sort of percentage bonus (e.g. Win=+10% to PSR adjustment, Loss = -10% to PSR adjustment) ... just not to the actual MS itself
4) As for thoughts on the MS calculation, I can't really say until I see a response to #1 listed above.
5) As for Matchmaker, I like the one I read (somewhere) that went back and forth from Team 1 and Team 2 using PSR, although I would want some sort of average team PSR calculation in an attempt to keep the "relative skill level" of both teams close.
Other thoughts:
A. For those of you rage-hating enemy LRMs ... bring AMS or ECM ... and actively search for/call out/shoot enemy UAVs. I've been in too many matches where someone calls out an enemy UAV and hardly anyone tries to take it down.
B. For those of you rage-hating that friendly AMS boats get too much MS bonus ... All I have to say to that is ... they kept you alive longer so you could do more damage, didn't they? If they didn't, then you probably should have tried to be closer to them (and yes, I know that isn't always possible). If you don't like them that much ... then I'll gladly have them stand close to me...thanks.
C. For those of you rage-hating those LRM boats that stay WAY back, like 600+ meters from the front line, tell them to move the hell up with the rest of the team or else there would be repercussions. There were three matches that I remember where there was this assault LRM boat that wouldn't move with the team...in one match, one of my teammates (I guess that he had dropped with the guy running the assault several times that night and was quite familiar with his 'antics') ... sent a message to the other team that there was a juicy LRM assault boat in so-and-so grid that refused to be a team player and was alone, ripe for the picking ... The enemy light mech thanked us for the meal... <g> ... The very next match, we got the assault LRMer again (remember what I said about losing the pug lottery?), the same teammate asked him whether he was going to stay with the team this time and was promptly told that "He would play his Assault mech the way he wanted to" ... my teammate's response to that was to take 1 leg as soon as the match started. The assault guy asked "Why'd you do that?" ... My teammate's response was: "What's the problem? You never move anyway..." ... I ROFL'd and damn near choked on my drink... Later that night, that same assault dude showed up again but this time he was on the opposing team... Me and another light took the long way around and went after him as I knew that he'd probably be way in the back, alone, and I was right ... we feasted ... slowly, as I had told the other light not to kill him quickly ... funniest thing was that no one from his own team ever showed up to help/rescue him. After he died, I told him that he'd still be alive if he had been with his teammates.
D. For those of you rage-hating those "guys-that-always-hang-at-long-range-not-sharing-their-armor-and-using-the-rest-of-you-as-meatshields" ... all you can really do is to 1) ask them to move up ... 2) have the rest of your team move to them ... or 3) if you're feeling rather malicious ... drop an arty on them ... (wait, did I just say that out loud? )
E. As for teamwork, it's on each and every one of us to try and herd the cats ... er ... PUGs/Potatoes/whatever you're calling them these days ... I've noticed that there's too many matches where everyone's just doing their own thing trying to kill enemy mechs ... but it'd go SOOO much better if someone/anyone called their target and got coordinated, focused fire on said target (I know ... I'm dreaming of the past ... I'm old ... remember?)
Anyway, for those of you who actually took the time to read this ... I thank you ...
Catch you guys on the battlefield (cause I spent WAY too much time reading this thread -AND- writing this response <g>)
-Airwolf-
Edited by Airwolf, 11 June 2020 - 01:05 AM.
#369
Posted 11 June 2020 - 01:48 AM
RCore, on 11 June 2020 - 12:17 AM, said:
About concerns of using MS for PSR, and the PSR becoming 'divergent' with the zero-sum proposed: If it happens, this would only affect the extreme ends of "below Tier 5" and "above Tier 1" players. Exceptional players like these will likely always exist. But note also, that if these exceptional players are matched against similarly exceptional players the gameplay would still be balanced - maybe these people can form the extreme top and bottom tiers. This leads to the other factor of what ranges of PSR form each Tier, which is another variable that can be adjusted.
(Edited for clarity.)
On your first paragraph, that is largely correct. Ratio-based ratings would trend towards 1.0. The matchmaker would effectively be trying to push you towards 1 by matching you against better or worse players depending on how far and in which direction you deviate from 1. Your success against those players would be the force pushing back against the matchmaker, until a balance is reached. I assume that the larger the playerbase was, the better the matchmaker would be at keeping people closer to the middle, but even smaller numerical deviations would remain accurate for ranking players.
On your second paragraph, I only did some very cursory analysis so I am not absolutely certain, but my main concern is in how your rate of movement away from the mean would depend on how many games you play. If there are periodic PSR resets, this would probably not have the time to really become an issue and how much of an issue it would be depends on how exactly PSR and tiers are used in matchmaking.
If tiers are unmoving ranges and the matchmaker doesn't know if you are barely in tier 1 or super-tier-1 rocketing towards infinity, this would probably not cause matchmaking issues. The distribution of players into tiers might end up being a bit weird if the ranges fail to match the range of actual player peformance tho.
If tiers are percentile based, then they'd be more accurate divisions of the population but PSR not converging would spread them away from the mean and potentially make it harder for those who play less to get out of Tier 3 to where they really belong, since the PSR of outlying players would depend on how many games more they have played.
In either case, if the matchmaker uses actual PSR numbers and not just tiers to make the teams, a divergent rating is going to get weird over time. For example, lets take two players we know are both extremely good, Lurm God and Bowser. Lets say Bowser has gotten really excited and played a thousand games after PSR reset while Lurm God has only bothered to play one hundred in the same time. Both are rocketing above and beyond Tier 1 but thanks to the volume of games played Bowser is at Tier 1 + two million while Lurm God is at Tier 1 + one million. If the matchmaker uses the raw numbers and thinks Bowser is twice as good as Lurm God, teams are going to get weird.
#370
Posted 11 June 2020 - 02:30 AM
2. Make PSR zero sum, 2-4 in the middle with no change is fine.
3. Rank purely by match score to award PSR changes.
For MM Assignment
4. 1/23/45/67/89/etc assignment based on PSR - PER CHASSIS VARIANT.
Start a new chassis = 100 matches all with player average PSR for either that CHASSIS or WEIGHT CLASS.
Start a new player = 100 matches, all with 3/4 of player base target PSR for the starting VARIANT.
5. Sum PSR for a group when assigning, top score + 1/8 of each additional player. Team 1 4 players = 1234, solo 5 goes to team 2, a second group goes to the team 2 678, + solo 9, back to team 1 for solo 10, etc.
Assignment is NOT by number of players on each team, but total PSR on each team until a team (probably #1) is filled. So a hotshot 4 man might get the next 7 ranked in the queue opposing them before they get their next player added to their team.
6. Monthly, move 5% of the difference between the "mid" PSR score, and your score for each VARIANT.
If you have to keep groups in the QP queue
7. Groups get Sum of group PSR change. Then ranked internal to the group. 4 man = top gets avg +2, bottom gets avg-2, middle get avg.
Edited by David Sumner, 11 June 2020 - 02:34 AM.
#371
Posted 11 June 2020 - 02:44 AM
Nightbird, on 10 June 2020 - 08:29 PM, said:
How would you handle teams of 2, 3, 4?
As far as how the MM organizes players from tiers, for all solo players matches this is already how it does it. The reason it doesn't work is because past MS is a poor predictor of performance on the team. (For the people that haven't hit the T1 cap, total past MS earned is their PSR bar location today.) It doesn't go by the 5 tiers but rather by a more granular number in the background. (The EXP bar line)
This is also why all the proposals based on MS won't work. Past MS is only loosely related to odds of winning future matches, whereas past WLR is strongly related. Picked any match score on Jarl's list (not adjusted) and look up and down the list at the player's WLR. You'll see a huge variance. This is clearer when you take all the data and graph it, showing a clear pattern that MS is inferior to WLR.
Rank players in order.
If a player is in a group, put all the players in at the top group player score.
Optional: Increase the PSR of group players by 5- 10% per additional group member.
Don't assign players 1,2,2,2,2,2 ... if there is a group.
Assign players such that the total PSR swaps over until one team is full.
Example
1,5,7,8 are a 4 man. 2,3 are a pair, rest are solo.
Team 1: 1,5,7,8 PSR totals 750
Team 2: 2,3 PST totals 700
Team 2: 4 PSR totals 760
Team 1: 6 PSR totals 810
Team 2: 9 PSR totals 800
Team 2: 10 PSR totals 835
etc.
Edited by David Sumner, 11 June 2020 - 02:45 AM.
#372
Posted 11 June 2020 - 02:58 AM
Nightbird, on 10 June 2020 - 02:19 PM, said:
Except.
Groups > Solo for wins. Yes?
So there will always be more Solos on the losing side, simply because more of the winning side will be in a group.
#373
Posted 11 June 2020 - 03:04 AM
Dauntless Blint, on 10 June 2020 - 11:06 AM, said:
Eg. How good is a guy at a Piranha against 10 other guys using the same mech then when he gets in an assault and pancakes at that archetype because he/she is a light inclined player. (more likely assault mains trying to light but the example makes sense I hope)
I'm just wondering how going from an archetype or specific mech your great at and trying to diversify to one your poor at could look like artificially lowering you true effective PSR rank in another mech.
Oh hell yes.
I can BARELY pilot a light in combat.
Because the view is updating faster than my GPU (or maybe my eyes) can handle it.
PSR should be VARIANT specific. Start a new variant? Inherit your average CHASSIS score. No chassis score? Inherit your average CLASS score. No class score? Inherit your average score. No player score? Inherit 3/4 of the "target" average score (because we'll assume you know nothing yet).
#374
Posted 11 June 2020 - 03:10 AM
Capt Deadpool, on 10 June 2020 - 01:04 PM, said:
I understand what you are saying, but no I don't really care at all. Utilize wins as the primary metric for PSR/matchmaking or else you are harming the teams you are on even if you are 'better than average' than others in some ineffective chassis. Wins keeps things simple and that is the entire purpose of the game: winning, not relative skill compared to others in the same chassis. People can make alt accounts for fun/ineffective mechs that don't get wins if they want.
If I am a thousand times better than an average player in my favorite crappy mech but don't get consist wins, then I should be content in whatever tier I end up in, i.e. a crappy player in a meta mech is many times more valuable to a team/winning than a good player in a crappy mech, and people need to practice in their chosen chassis not just to be better than others in the same chassis, but as effective or better as players in effective chassis that get wins.
All of which is actually an argument in favour of a PSR per mech variant.
#375
Posted 11 June 2020 - 03:18 AM
Wraith of Shadow, on 10 June 2020 - 09:56 AM, said:
Also kind of want dead guys chatting with the living to be cut, both for the drama of your voice dissolving into static as you "die" but also because it's kind of odd for dead guys to give advice.
ie 2 chat channels, 1 for the living and the other for the dead.
That's probably beyond the scope of this discussion though.
But often my greatest contribution to a match is being able to focus and call targets better after my mech dies.
It's almost like a team benefits from having a commander or observer that's not currently being shot at.
#376
Posted 11 June 2020 - 03:27 AM
My think is: Something like pool system, non zero sum (chart below)
1. In theory every player can have maximum 36 points.
2. Every action give you a point ONLY ONCE. So you get point, when you kill a first mech. That is yours +1, but you don't get point for another kill. You get a point when you do first scouting, another +1, and no point for another scouting. So every player has his own pool of points, and that is all your matchscore. So matchscore don't go like 300-400 points. Maximum matchscore is 36, in theory from the chart. (But it's impossible to get 36, one of obvious reasons: you will not get points for capture in skirmish mode etc). Note also this, with sucha system match in skirmish will give all players less points. But in Domination, Incursion etc .players have chance to gain more points (famous "going by the objectives"). Of course all actions from charts still give C-bills and experience points through whole match.
3. Base value for loosing is high -10. But try to earn +10 in that system. First, it's looks like easy task. But "one point one time only" means you have to do many different things to collect that +10. Loosing almost always means dying, so it's like -15. But still you have only +1 for a win. Let say it's only +1. But very important +1
Examples
Player A has 20 points (I think it's pretty good score in that system) at the end of the match. But he lost that match (-10) and died (-5). His final score is +5. So he is going 5 points up in Tier system.
Player B wanted to be sneaky, he scored with damage done, spotting assist, component destroyed, and kill assist. (+4 in total) .Then he decided he don't wanna play anymore (or made a mistake), he rushed at the enemy team and gave them easy kill. (For him it's -5, for guys, who were shooting him it's at least 2,3 points for all of them. Of course only then, when those points are their "first points". Team of Player B won that match, so he get +1 more. But still his final score is 0. If he want better score, he has to do many other things to get points. And let say that team of Player B lost that match. Then Player B final score would be -10. So he would go down -10 points in Tier system.
I see it as a way to simplify things. Don't think all values are correct, and maybe "one point one time only" is not the best idea. But it's simple. Maybe this solution also needs some other actions you can do to get points or lost points? It also opens a way to add other actions for gaining and loosing points or do some changes. And tier bars must work fine with whatever numbers will be implemented here.
Problem of that. Many "+" to go, only few "–" to lose points. But:
Player A a died quick, so he hadn't got any chances to go up with points. Then he fnished match with -5. That is why "penalty" for dying should be in that chart.
By changing values from the chart we can make game more focused on this or that, in terms of balancing this.
In Faction Play this goes by one pool of points for every mech (drop). So I can go like +10 in first mech, -5 in second and so on. After the match you get a sum of your mechs points. So you can go like 100 and more points in one match in FP. Maybe some bonuses for alive mech, like +1 point or so? Then, to make it funny, we can make an overall ranking of players/ unit in FP based on simple scoring system.
PS. I have no idea how can it go with matchmaker.
EDIT: Tried to simply copy paste chart, but it looks this way, I will try to link it somehow. Basically almost all actions for gaining points are +1.
Alright, here is link https://docs.google....dit?usp=sharing
Edited by Bistrorider, 11 June 2020 - 03:58 AM.
#378
Posted 11 June 2020 - 03:38 AM
Ridir Semii, on 09 June 2020 - 01:43 PM, said:
I'm already lunch for them.
It's impossible, even with max turn boosts, to actually keep them in sight long enough to hit with more than a glancing laser shot.
When (not if) the rest of the side runs off and leaves you on your own (which is what actually happens, rather than my "standing at the back" by choice), you're toast for the first light that comes along.
#379
Posted 11 June 2020 - 04:02 AM
Andrzej Lechrenski, on 10 June 2020 - 05:15 PM, said:
I don't know if I agree that is how standard deviations work. Using standard deviations also won't help very much with such a low population, as it essentially means that either the extremes will never get matches, or else the MM will be loosened so much that it was a wasted effort to begin with. Might as well just go with a percentage based system where top W/L is 99.9999+% and bottom is 0.0001% and just let them loosen valves from there.
Considering using only the most recent hundred or so matches is a fine idea, though.
Well as I understood it, Sd would be used to sort the pilots into tiers. The pilot's tier would be information for MM on the pilot's general ability.
The tier system itself would not be strict. MM will still need to compose teams of pilots from multiple tiers. This is due to the low population. The main idea is however to find balance between both team's chances to win. This is why we are even discussing PSR, tiers and MS. The point is to find such a metric that will as accurately as possible represent the pilot's skill.
WLR is one proposition. I don't particularly like it tbf. Over time a perdectly balanced MM would bring basically everyone to WLR of 1.0
The other metric is MS which I find better as it's more robust when faced with "perfect MM". But I admitted already that tweaks are needed for it to accurately represent actual skill.
Oh and yes, MS based PSR should creep back to "neutral" value over time to curb divergence. Be it a limit on match number taken into account, basing PSR on current and one/two previous seasons or an actual real-time "tick".
#380
Posted 11 June 2020 - 04:19 AM
David Sumner, on 11 June 2020 - 02:58 AM, said:
Except.
Groups > Solo for wins. Yes?
So there will always be more Solos on the losing side, simply because more of the winning side will be in a group.
@David Sumner: While the number of matches I observed from my screenshots was not large, it appeared that this is not true... A group of 4 is (within my limited sample) more likely to win vs 2+3 and a group of 3 against a 2+2. So the winning side (at least in my limited sample) is likely to contain more solos on the winning side, IF both sides have groups. Also, there were numerous examples of cases, where 12 vs 2+10 ended in the 12-side victory (in other words, some pairs are not good enough to sway matches by themselves).
3 user(s) are reading this topic
0 members, 3 guests, 0 anonymous users