63 replies to this topic

[evilauthor]

I think the problem with the current PSR system is that its data comes from a player's ENTIRE history of playing the game. But the thing is, people (theoretically) get more skilled with time. The more you play, the better you get at the game (again, theoretically). So even if you're a good player now, you have your early history of bad play weighing you down. And of course different people learn at different rate.

So I think a better way to evaluate a player's skill is to use their match scores from the last X number of matches that they've played with each new match played replacing the stat of the oldest match in the list. This gives the system a better feel of how skilled the player currently is. Of course, the X value aka the number of matches used for evaluation has to be broad enough to account for the bad luck and good luck when it comes to team selection.

[Bombast]

yrrot, on 08 April 2018 - 07:08 AM, said:

Part of me wonders if it would be better to try to balance teams based on the players Solaris 1v1 Elo for the division of mech they are in for QP, if PGI wanted to go to an Elo-based system there again.

They said that's something they will be looking into post-Solaris release.

They obviously wont be able to use a players actual Solaris Elo ranking for QP though, since many (I'm going to guess most) players will not be playing Solaris. But they want to bring the math to QP stats. Somehow. It's worth a shot, but it sounds like it can go south fast.

evilauthor, on 08 April 2018 - 07:22 AM, said:

I think the problem with the current PSR system is that its data comes from a player's ENTIRE history of playing the game. But the thing is, people (theoretically) get more skilled with time. The more you play, the better you get at the game (again, theoretically). So even if you're a good player now, you have your early history of bad play weighing you down. And of course different people learn at different rate.

PSR is basically accumulative though. Beyond your first 20-50 games or so, games essentially treats every match as a -0.5, 0, +0.5, or +1.0 to PSR (Not literally those numbers, but you get my meaning). There's no averaging with earlier matches that drags you down - The better you play, the faster you rise in PSR, no matter what you've done before (Of course, playing badly makes your PSR go up too, but thats a separate issue)

Now, Jarl's list's 'Overall' stats leave an anchor around your neck, but the rankings do count newer matches more than older matches.

Edited by Bombast, 08 April 2018 - 07:30 AM.

[Shifty McSwift]

What if they inverted the tier direction so that it counted up as you went up in tier and gave it no maximum, only an increasing gap for tiering up to the next tier? Then people should start leveling out at a certain point once they get placed with the higher numbers?

Or better yet what if the rewards system functioned vastly/slightly differently and more practically for promoting team play? For example inverting the damage to reward system, so that you basically pay a player when dealing damage to him (still reward takedowns and assists/KMDDs too to some degree), as this would reward accuracy, as well as rewarding good tanking/good usage of team armour.

What if a lot of things really, could play that game all day.

[DAYLEET]

If you guys want skill based match there will be Solaris. Though you wont be able to blame anyone else but you in Solaris so think hard before you get in.

I said it already. ELO in a 12v12 pub queue is genuinely stupid.

[El Bandito]

evilauthor, on 08 April 2018 - 07:22 AM, said:

So I think a better way to evaluate a player's skill is to use their match scores from the last X number of matches that they've played with each new match played replacing the stat of the oldest match in the list. This gives the system a better feel of how skilled the player currently is.

We can already check the stats of the latest seasons for every player.

[Snowbluff]

El Bandito, on 08 April 2018 - 08:47 AM, said:

We can already check the stats of the latest seasons for every player.

No, let's pretend season 21 didn't happen for me... I suck when chasing events

[Zergling]

MrMadguy, on 08 April 2018 - 04:55 AM, said:

What I see - is that there no correlation between WLR and personal performance.

As you can see, correlation is way too weak:

If you think there is no correlation between MS and W/L in your own stats, or that graph shows a weak correlation, then you really don't understand statistics.

Edited by Zergling, 08 April 2018 - 06:38 PM.

[Nightbird]

That graph is not a weak correlation, it probably has a very low p-value.

PGI just needs to hire someone who knows math to design the match maker. Find a statistical model that minimizes final score differences.

[Deathlike]

All I can say is... if you think you're good, and then we plug into those numbers... you're going to have a violent reaction from the community.

If you're barely holding a 1 W-L ratio, you are barely holding your own.

Even in the death squads that rack up the wins, you still actually have to produce some value to the team. On any extended losing streaks, you're probably not doing enough (aka carrying) to make a real difference.

It is far easier to notice how (badly) you are losing compared to winning streaks where you may have just been a spectator to the destruction of an opfor.

PSR is flawed, but ignoring any sort of correlation to W-L from some combination of K-D, damage, and match score probably means you're "just another guy" and not so much a "credit to team".

[Asym]

If the current system isn't completely working; and, ELO didn't work; I'm not sure any system would work at all because the game has too many variables, too many mech platforms, to many exceptions and nothing, like % of shot hits or accuraccy percentages that are finite enough to measure competence.

Having said that, mybe the answer goes in the other direction: lack of precision... Maybe, don't use a match maker and only arrange matches by mech weight class and everything else is random..... Anyway you do this, a lot of people will complain.

And for heavens sake, leave SOLARIS out of everything except Solaris...... Good grief...

[Deathlike]

Asym, on 08 April 2018 - 07:12 PM, said:

If the current system isn't completely working; and, ELO didn't work; I'm not sure any system would work at all because the game has too many variables, too many mech platforms, to many exceptions and nothing, like % of shot hits or accuraccy percentages that are finite enough to measure competence.

Having said that, mybe the answer goes in the other direction: lack of precision... Maybe, don't use a match maker and only arrange matches by mech weight class and everything else is random..... Anyway you do this, a lot of people will complain.

And for heavens sake, leave SOLARIS out of everything except Solaris...... Good grief...

The problem with Elo has more to do with population and actual skill distribution.

I believe that at most there are 20% of the overall population is actually far and away better than the playerbase. In fact, the average player is actually not that competent. It was actually easy to see how the best players had a problem making a match based on such criteria - the Elo distribution is slanted in such a way preventing an "even distribution". You either are "meh" or "above the brown sea" level of skill. It's actually pretty sad to describe it in such a manner. This ignores mechs altogether though. Picking your mech only amplifies or handicaps your potential. So the reality is that a good player in a bad mech can make a bad mech look good, but on the other hand a bad player in a good mech can still produce pretty terribad results. It's not rocket science here.

[justcallme A S H]

MrMadguy, on 08 April 2018 - 02:28 AM, said:

My stats - are best proof, that WLR is completely meaningless for pure QP player.

Meaningless? Wut?

Here are MINE and I basically play only SoloQ as well. Except season 18, that was basically all GroupQ.

WLR is a good indicator.

My carry power is about 8 in YoloQ and it's funny, many know this when they see me on the otherside... I often get chased/dived on - it's good fun

Edited by justcallme A S H, 08 April 2018 - 07:23 PM.

[Zergling]

justcallme A S H, on 08 April 2018 - 07:21 PM, said:

My carry power is about 8 in YoloQ and it's funny, many know this when they see me on the otherside... I often get chased/dived on - it's good fun

Yup, I intentionally focused on killing you when I saw you on the enemy team last week. My Viper would have been unlikely to have done that solo though; I needed the help of teammates to bring you down.

Edited by Zergling, 08 April 2018 - 07:29 PM.

[justcallme A S H]

Deathlike, on 08 April 2018 - 07:17 PM, said:

The problem with Elo has more to do with population and actual skill distribution.

I believe that at most there are 20% of the overall population is actually far and away better than the playerbase. In fact, the average player is actually not that competent. It was actually easy to see how the best players had a problem making a match based on such criteria - the Elo distribution is slanted in such a way preventing an "even distribution". You either are "meh" or "above the brown sea" level of skill. It's actually pretty sad to describe it in such a manner. This ignores mechs altogether though. Picking your mech only amplifies or handicaps your potential. So the reality is that a good player in a bad mech can make a bad mech look good, but on the other hand a bad player in a good mech can still produce pretty terribad results. It's not rocket science here.

Correct.

I'd say about 10% of the population TBH. About 1,000 players at most are "that" good. And then there is the 1% right at the top that are far and above all of us and those are the guys @ Mechcon in the top 3 teams etc

[Deathlike]

justcallme A S H, on 08 April 2018 - 07:21 PM, said:

Meaningless? Wut?

Here are MINE and I basically play only SoloQ as well. Except season 18, that was basically all GroupQ.

WLR is a good indicator.

My carry power is about 8 in YoloQ and it's funny, many know this when they see me on the otherside... I often get chased/dived on - it's good fun

I always find that those that consistently struggle have always hit the proverbial wall. Maybe they don't know how to get better. Maybe they don't want to get better. Sometimes watching the really good players is how you "steal" or learn concepts that could actually be used... but at this point with certain people, it's just a hackusation instead of trying to be observant and learn something.... whether it be solo or group queue.

The potential is always there to unlock. It's a matter of whether you want to find the key to success. I guess it's better just taking a dump on players that are good instead of realizing how much better one could be. /shrugs

[justcallme A S H]

Zergling, on 08 April 2018 - 07:24 PM, said:

Yup, I intentionally focused on killing you when I saw you on the enemy team last week. My Viper would have been unlikely to have done that solo though; I needed the help of teammates to bring you down.

You and every other person out there lol

I get used to it. I actually have fun when it's 2-3 on one, try take down as much as I can.

I'll still rage after it because usually to dive on me it's obvious but the team is... Oblivious even if you tell them 5 times over VOIP.

The trials and tribulations of YoloQ. Fun/Challenging and frustrating at the same time.

[Zergling]

justcallme A S H, on 08 April 2018 - 07:28 PM, said:

You and every other person out there lol

I get used to it. I actually have fun when it's 2-3 on one, try take down as much as I can.

I'll still rage after it because usually to dive on me it's obvious but the team is... Oblivious even if you tell them 5 times over VOIP.

The trials and tribulations of YoloQ. Fun/Challenging and frustrating at the same time.

It sure didn't help that your 2 nearest teammates were a LRM boat Kodiak and a LRM boat Sun Spider, while the rest were completely oblivious.

[Deathlike]

justcallme A S H, on 08 April 2018 - 07:25 PM, said:

Correct.

I'd say about 10% of the population TBH. About 1,000 players at most are "that" good. And then there is the 1% right at the top that are far and above all of us and those are the guys @ Mechcon in the top 3 teams etc

Even I know it's unrealistic to be the top 1%. I don't even have the time for that nowadays.

Still, if I'm still able to perform 80% of their level - I feel that's a worthy achievement.

Then again, I see garbage play 99% of the time, so sometimes the effort of learning gets lost inevitably.

Free advice to get better? No. Call others out for doing better thru teamwork or other supposed magic that happens? That's the MWO ticket!

Edited by Deathlike, 08 April 2018 - 07:32 PM.

[Samedi Wretch]

MrMadguy, on 08 April 2018 - 04:55 AM, said:

What I see - is that there no correlation between WLR and personal performance.

As you can see, correlation is way too weak:

Your data contradicts you. An R^2 of .25 is pretty ******* large actually. That means 25% of the variance in wlr is accounted for by variance in match score. Learn to stat. ;P

And guess what -- winning is a pretty reliable indicator of a winner.

Zergling, on 08 April 2018 - 06:37 PM, said:

If you think there is no correlation between MS and W/L in your own stats, or that graph shows a weak correlation, then you really don't understand statistics.

Ahh zergling beat me to it.

Edited by samadhiVOID, 08 April 2018 - 08:10 PM.

[0Jiggs0]

If W/L is to be used to determine matching, then PSR should not increase or decrease at a 50% win rate. A 50% win rate would indicate the player is competing against others of similar skill, a higher percentage would indicate opponents are too weak, and so on.

However, W/L in Quick Play is often not a matter of individual skill, as the team's collective skill is frequently the determining factor. An AFK, disconnected or simply incompetent player can easily cost their team a victory, so basing PSR on win rate seems counter intuitive to me. I believe basing PSR on relative-to-team Match Score would be a more effective assessment of player skill and result in better matchmaking overall. Players who trend in the top percentiles among their teammates would move up until their performance normalizes, with an ideally ranked player scoring in the top, middle, or bottom percentile of their team's average score with equal frequency.

(removed duplicate text and a typo)

Edited by 0Jiggs0, 08 April 2018 - 10:24 PM.