139 replies to this topic

[Jay Leon Hart]

Xavori, on 28 November 2017 - 11:08 AM, said:

I don't have to prove the leaderboard is random. You already know the matchmaker is random. Nobody argues that PSR is not a meaningless rating, and since the matchmaker barely uses that anyway, you get random matches. I mean, you're not seriously suggesting I'm wrong about this?

So if the matches are random, how then do you think you get not random results?

No, you don't have to prove anything. You can be as wrong as you like, it's fine.

If the results are demonstrably not random, how can you claim they are?

Somewhat of a tangent, IIRC the matchmaker in TitanFall 2 tries to get everyone to have a W/L of 1. To me, this is how a matchmaker should work. You are playing with people who, together, give each other a 50/50 chance to win or lose.

Do we know what, if anything, PSR attempts to trend towards?

[MischiefSC]

Xavori, on 28 November 2017 - 11:05 AM, said:

When I keep telling you the leaderboard is garbage, why do you keep trying to insist that I use the leaderboard as part of my argument? I'm not going to because the leaderboard is garbage. The only thing you can pull from looking at the leaderboard is which players care about which stats and which players either get lucky teams more often or play with good teammates more often.

If the leaderboard is garbage than the results would be random. They would swing randomly from one side to the other for everyone every month.

Please show that.

Arthur C. Clarke once said that to a primitive culture any sufficiently advanced technology is indistinguishable from magic. You are equating some basic mathematical principles with magic.

If the leaderboard is garbage then show me the random swings in the leaderboard that make it garbage.

Jay Leon Hart, on 28 November 2017 - 11:13 AM, said:

Do we know what, if anything, PSR attempts to trend towards?

Tier 1.

Which is why it's a fundamental failure.

[Xavori]

Jay Leon Hart, on 28 November 2017 - 11:13 AM, said:

No, you don't have to prove anything. You can be as wrong as you like, it's fine.

If the results are demonstrably not random, how can you claim they are?

Somewhat of a tangent, IIRC the matchmaker in TitanFall 2 tries to get everyone to have a W/L of 1. To me, this is how a matchmaker should work. You are playing with people who, together, give each other a 50/50 chance to win or lose.

Do we know what, if anything, PSR attempts to trend towards?

PSR wouldn't be able to pull of Titanfall 2's matchmaker because you cannot have just 5 tiers. Not that it matters anyway because PSR for everyone in MWO will pretty much go up over time because it's easier to go up than down. I made it a point to hold Tier 3 for a long time, and I can assure you, you really have to work at it (like run lights all the time, don't do a lot of damage, and so on)

As for the leaderboard since you really want to talk about it...

There is no separation between group and solo queue. Given the interdependence on teammates, players who play with good teammates absolutely will win more than those who play with friends, or just people they meet in LFG, or casual players, and so on. I recognize quite a few names from the top of the leaderboard as players who never drop solo, so it's not at all surprising to see them near the top. So I'm not going to be able to use any of them because of course they're going to show consistent results.

That means to comply with your request, I'd have to limit myself to players who drop solo queue exclusively. I don't have any idea if they even exist. I mean, I drop solo a lot, but I also drop with my unit a lot. And even in unit drops, we do silly things like Urbie rushes or run full 12 man and just eat the massive tonnage penalty and so on. And I'm pretty sure that's pretty commonplace in a lot of units.

So no, I can't use the leaderboard to comply with your request. It's garbage. There are literally no meaningful stats to be found on it because your stats are so tied to team performance, and teams are so very random in solo queue, and team quality isn't considered in group queue.

[MischiefSC]

Xavori, Asym, look at the leaderboard fo the last 3 months.

Take the total number of matches you WON for the season, divide it by the total number of matches played.

That's the percent of matches you won for the season.

It will almost certainly be within 3% one season to the next.

See how it doesn't swing wildly, like it would with a random distribution?

I almost never play group queue. I haven't had 10 matches in the group queue in any given season I've ever played in this game. Do the same with mine, do the same with anyones.

So you have absolutely no data to back up what you're saying. I do; you can use the link you posted to me for TrueSkill to see the basic math principles behind solving for a players value out of a group. You can also google statistics and algebra and percentages. Then you can look at the leaderboard and see consistent results for people month after month.

At this point you have flatly refused to show any evidence for anything you've said - which isn't surprising because what you're saying is that math doesn't work. You're also trying to say that the population in group queue is as big as the population in QP, which is clearly and demonstratively wrong. While we would all agree that mix of group and solo queue stats are bad and Tarogato has requested they be split for the vast majority of the games population they are irrelevant.

Please provide any mathematical principle to support your position or any actual data points. You've provided a link to data on TrueSkill that clearly states and shows both in description and via mathematical formula that it's derived from win/loss.

One more time, if win/loss on the leaderboard was random then results would be random. Show random results in the leaderboard stats please. IF they are garbage please show examples of them being garbage. In the stats. Show garbage stats please. You know, wild, unreliable, random swings for the same person season after season.

Edited by MischiefSC, 28 November 2017 - 11:31 AM.

[Xavori]

MischiefSC, on 28 November 2017 - 11:17 AM, said:

If the leaderboard is garbage than the results would be random. They would swing randomly from one side to the other for everyone every month.

Please show that.

Arthur C. Clarke once said that to a primitive culture any sufficiently advanced technology is indistinguishable from magic. You are equating some basic mathematical principles with magic.

If the leaderboard is garbage then show me the random swings in the leaderboard that make it garbage.



Tier 1.

Which is why it's a fundamental failure.

Dude, I don't consider math magical. I totally understand it. The problem you and I are having is that you don't. You are trying to apply things you know, statistics, to numbers that don't fit the prerequisites. You even acknowledge that last bit by acknowledging that PSR is a failure.

So if PSR is a failure, and the matchmaker is creating teams sorta kinda trying to use it but giving up because of the small player base, how then do you think I'm wrong about matches just being random? And if the quality of teams is random, how can you possibly try to come up with any meaning ascribed to the numbers that come out of random matches?

Go back to the TrueSkill you keep bringing up. It builds in the quality of the match, ie. not random. It has to. It considers the skill of the two sides when determining how much to move the winners up and the losers down. It considers a lot more than that as well. It works because the matchmaker isn't just throwing random teams together, but instead is trying to get as close to equal rating on both sides as it can. And over time, if you keep doing that, you do actually get meaningful numbers because those numbers aren't built on random numbers.

And perhaps most of all, why do you keep thinking W/L is going matter? It will be ~1 in a good matchmaker setting and random in a bad one.

[MischiefSC]

Xavori, on 28 November 2017 - 11:35 AM, said:

Dude, I don't consider math magical. I totally understand it. The problem you and I are having is that you don't. You are trying to apply things you know, statistics, to numbers that don't fit the prerequisites. You even acknowledge that last bit by acknowledging that PSR is a failure.

So if PSR is a failure, and the matchmaker is creating teams sorta kinda trying to use it but giving up because of the small player base, how then do you think I'm wrong about matches just being random? And if the quality of teams is random, how can you possibly try to come up with any meaning ascribed to the numbers that come out of random matches?

Go back to the TrueSkill you keep bringing up. It builds in the quality of the match, ie. not random. It has to. It considers the skill of the two sides when determining how much to move the winners up and the losers down. It considers a lot more than that as well. It works because the matchmaker isn't just throwing random teams together, but instead is trying to get as close to equal rating on both sides as it can. And over time, if you keep doing that, you do actually get meaningful numbers because those numbers aren't built on random numbers.

And perhaps most of all, why do you keep thinking W/L is going matter? It will be ~1 in a good matchmaker setting and random in a bad one.

Then show your math please.

You also didn't read anything in the TrueSkill setting -

Your w/l will not always be 1.0. Good players will be above 1.0, bad players below it. W/L is not the same as your TrueSkill score - it goes into that on the page you linked me but absolutely clearly did not read.

TrueSkill, and Elo, and any matchmaker for that matter, gives you a score, a ranking based off your W/L. Then it adjusts that score up or down based on who you've beaten and lost against. So in TrueSkill a great player may have an 8.0 W/L ranking - however if he's not playing against and beating other highly ranked players his TrueSkill ranking will only be moderately high.

It goes over that in detail in the link you yourself put up.

If w/l is about 1.0 in a good matchmaker then for the the matchmaker is good I guess? You're within a few percentage points of a 50% win rate. Your win/loss is not random.

The only reason MWO doesn't need a TrueSkill class matchmaker is we don't need a ladder ranking - just distributions, to make matches. We don't have enough population to really warrant a ladder system. TrueSkill gives more detail than we need. It also has a lot of layers of complexity that would have no value here as it's designed to equate your performance in 4 v 4 or even 12 person FFA with your 12 v 12 stats to give you a consistent ranking. We don't need that.

Again, show me the data in the leaderboard that is garbage and unreliable. Show examples where it's random.

Also, the entire point of statistical analysis is taking numbers you do know and using them to figure out the numbers you don't. In fact that's part of basic algebra. Solve for X.

You also never answered my question about two drivers - it's really, really simple. I'll re-quote it for you.

Quote

So you have two people. One guy, Bob, texts while he drives, pays little attention to what's going on around him. He runs red lights and drives a beaten up old car with no seat belt or safety features.

Another guy, Fred, is a very safe driver. Pays attention, drives carefully, drives a new car with all the best safety features.

They both have a 20 mile commute through the same traffic 5 days a week.

Is one of them more likely to get in an accident than the other? In an accident is one more likely to be seriously injured?

Why?

Edited by MischiefSC, 28 November 2017 - 11:49 AM.

[Jay Leon Hart]

Xavori, on 28 November 2017 - 11:24 AM, said:

So no, I can't use the leaderboard to comply with your request. It's garbage. There are literally no meaningful stats to be found on it because your stats are so tied to team performance, and teams are so very random in solo queue, and team quality isn't considered in group queue.

OK, I'll pull some stats from a few players later/tomorrow.

Since they're apparently random, it shouldn't even matter which ones.

[MischiefSC]

Another point you keep missing -

Yes, the current matchmaker is bad. However that doesn't affect its impact on your win/loss other than that it isn't doing as much as it should to keep players at different skill levels playing each other. Your win/loss is still accurate and relative to every other player, just spread more on a Gaussian curve and distribution than a Logistic one.

The challenges you face are the same as the challenges everyone else faces. How well you win in spite of them and how often you can help your team win in spite of them directly reflects in your win/loss. So players who are better at the game will win more than others, consistently.

That's Law of Large Numbers. You're stuck in the gamblers fallacy - you're attributing short term streaks of win or loss, good and bad luck as being representative of the average and it's not.

https://en.wikipedia...f_large_numbers

Read that.

[Xavori]

MischiefSC, on 28 November 2017 - 12:07 PM, said:

Another point you keep missing -

Yes, the current matchmaker is bad. However that doesn't affect its impact on your win/loss other than that it isn't doing as much as it should to keep players at different skill levels playing each other. Your win/loss is still accurate and relative to every other player, just spread more on a Gaussian curve and distribution than a Logistic one.

The challenges you face are the same as the challenges everyone else faces. How well you win in spite of them and how often you can help your team win in spite of them directly reflects in your win/loss. So players who are better at the game will win more than others, consistently.

That's Law of Large Numbers. You're stuck in the gamblers fallacy - you're attributing short term streaks of win or loss, good and bad luck as being representative of the average and it's not.

https://en.wikipedia...f_large_numbers

Read that.

Dude, I know Law of Large Numbers and Gamblers Fallacy. I'm not attributing anything to anything. That's the point. You're the one trying to attribute meaning to the leaderboard in spite of acknowledging it's based on random numbers that you can't know anything about.

If you want to apply the Law of Large Numbers...well...what's your range of numbers? You have to have one because you cannot have an expected value without a known range. And you've already acknowledged that PSR is garbage, and the matchmaker is garbage...soooooo.....???

It's interesting, isn't it that when you bring up TrueSkill, I point out that it specifically invalidates your desire to read meaning into wins and win/loss ration because it's ultimately designed to move w/l ration to 1 (side note: 1 is the expected value so the law of large numbers says you'll eventually get there after you first get matches between equally skilled opponents as the norm.) So you give up on it and pull out The Law of Large Numbers.

And now I'm pointing out that you cannot apply the Law of Large Numbers because you don't have an expected value, you don't have a range of numbers to give you an idea of necessary sample size to be "large" and you don't even know if the player skill levels of MWO's current player base fits a normalized distribution. So guess what? The Law of Large Numbers does not apply.

Ya know, it's almost as if I've studied a whole bunch of mathematics....

I cannot stress enough that the leaderboard is garbage.

And the entire point of these discussions I keep having is that it doesn't have to be this way. PGI could easily implement any one of many player ranking systems that over time would produce quality matches which in turn would produce a quality leaderboard. But one of the most important features of such a leaderboard would be that W/L ratio, that stupid number you keep trying to magically assign meaning to in the current PSR/random matchmaker created leaderboard, would be utterly and completely meaningless for player skill because every single player, no matter what their actual skill in playing the game, would have a W/R ratio that of ~1.

[MischiefSC]

Xavori, on 28 November 2017 - 12:38 PM, said:

Dude, I know Law of Large Numbers and Gamblers Fallacy. I'm not attributing anything to anything. That's the point. You're the one trying to attribute meaning to the leaderboard in spite of acknowledging it's based on random numbers that you can't know anything about.

If you want to apply the Law of Large Numbers...well...what's your range of numbers? You have to have one because you cannot have an expected value without a known range. And you've already acknowledged that PSR is garbage, and the matchmaker is garbage...soooooo.....???

It's interesting, isn't it that when you bring up TrueSkill, I point out that it specifically invalidates your desire to read meaning into wins and win/loss ration because it's ultimately designed to move w/l ration to 1 (side note: 1 is the expected value so the law of large numbers says you'll eventually get there after you first get matches between equally skilled opponents as the norm.) So you give up on it and pull out The Law of Large Numbers.

And now I'm pointing out that you cannot apply the Law of Large Numbers because you don't have an expected value, you don't have a range of numbers to give you an idea of necessary sample size to be "large" and you don't even know if the player skill levels of MWO's current player base fits a normalized distribution. So guess what? The Law of Large Numbers does not apply.

Ya know, it's almost as if I've studied a whole bunch of mathematics....

I cannot stress enough that the leaderboard is garbage.

And the entire point of these discussions I keep having is that it doesn't have to be this way. PGI could easily implement any one of many player ranking systems that over time would produce quality matches which in turn would produce a quality leaderboard. But one of the most important features of such a leaderboard would be that W/L ratio, that stupid number you keep trying to magically assign meaning to in the current PSR/random matchmaker created leaderboard, would be utterly and completely meaningless for player skill because every single player, no matter what their actual skill in playing the game, would have a W/R ratio that of ~1.

Except it's not random. That's the whole point you refuse to get. Player skill is not an infinite curve. No human behavior is an infinite curve or truly random. It's not even that big a curve or that wide of a spread.

Page 14 in the link you provided for TrueSkill.

You also seem to absolutely and totally not understand how matchmakers and win/loss work. Neither Elo nor TrueSkill push a player to 1.0 w/l. That's why you have a matchmaker with a different ranking instead of just using win/loss.

It assigns you a value and then modifies that value based on who you played against - so beating a player who's better than you raises you a lot, someone who's your skill a tiny bit, someone less than you not at all. So your win/loss only every changes by 1 but your ranking changes by a variable number based on your team and the other team.

Now you're trying to use terms you clearly don't understand.

The population of MWO fits a normalized distribution - every group fits a normalized distribution, you normalize the distribution based on the group. Again, the link you posted to the TrueSkill breakdown discusses this and how it groups and distributes and assigns ranks.

LLN works with infinite variance, however your variance in this is absolutely finite. You are 1/12th of your team and you're playing against a team of 12 every match. There are 11 variables on your team and you can either treat the other team as one averaged variable or 12 individual variables coming to an average - the results are the same.

The value of each of those variables is ranked along whatever skill curve you want; as detailed as you want to get however they're going to fall within a set range.

Because this equation holds true for every player everyone is ranking along the same curve. Because win/loss is zero sum, as in every single match there are 12 wins and 12 losses, every single players win/loss is relative.

There is no magic here save that you really, really want to pretend that your win/loss is not a direct result of how well you play, just like every other players is. You want to believe that you have no measurable impact on it. You are arguing that because teams in QP are random you have 0 impact on your team.

Take 12. Divide it by 1. You get 8.333%. That is your impact, every single match, on your team. How much value your 8.333% adds and sways the probability of your win by is a reflection of how well you play on average.

Again, the math you're refusing to get here is 1/12 and that 8.333% > 0%. So all the leaderboard does, by principle of basic mathematics, is identify in a 0 sum environment what your relative 8.333% is approximately worth on average relative to everyone else.

If you have a 49% win/loss then someone else has to have a 51% win/loss. Because that match you were in had 12 wins and 12 losses.

How much you carry your 8.333% of your team influences how likely your team is to win. A person who disconnects every single match is not going to win as often as the comp tier player in top meta on average, right? One of them is contributing more to his team, every team he plays on, every match, right? His average wins will be higher.

This is what you refuse to get. The matchmaker needs based off win/loss. While the current matchmaker does a poor job of building teams of players with a relative skill level that influences the enjoyment of the matches for the players, not the reliability of the results. Good players will win more than bad players on average in a QP environment. They will have a higher win/loss. Every high win/loss is matched by a low win/loss. 12 v 12, 12 wins, 12 losses, it all has to sum to 0.

1/12. 8.333% > 0%. Your win/loss is a reflection of how good you are at helping your team win games. As such it's the only useful metric to use for building a matchmaker. Hence why Elo and TrueSkill use win/loss as the metric they use to create a score from.

Once again - if the results of the Leaderboard are random then they would have a random distribution. From +100 to 0.01 range of variance for each player every single month. Show me that in the leaderboard.

Edited by MischiefSC, 28 November 2017 - 01:14 PM.

[Asym]

OK, let's call a truce... The forum is not a contest. There are no winners or losers.

And, most importantly, this entire discussion is MOOT...... There is no way PGI will or would even take on PSR or W/L......

So, let's call a cease fire and if you are out there one night playing, invite each other to drop and talk.....

How's that?

[MischiefSC]

Asym, on 28 November 2017 - 01:17 PM, said:

OK, let's call a truce... The forum is not a contest. There are no winners or losers.

And, most importantly, this entire discussion is MOOT...... There is no way PGI will or would even take on PSR or W/L......

So, let's call a cease fire and if you are out there one night playing, invite each other to drop and talk.....

How's that?

Here's the thing.

We used to have an w/l based matchmaker but people raged that it 'trapped' them. Essentially people believe that if they could just get on good teams they would get carried to wins more often. That resulted in PSR, designed to push people up into matches with better players. They also felt that the Matchmaker was trying to 'cheat' them by forcing them to a 1.0 w/l, when they felt they should be winning more often.

That didn't work, all it did was add mediocres and bads into gameplay with good players, resulting in the bads getting farmed. Most matches in T1 are more like an 8 v 8 or 6 v 6 with farmables filling in. This is frustrating for everyone. The bads get thrown up to T1 where they settle into a sub 1.0 w/l and get destroyed match after match after match, rarely getting kills or decent damage. The good players have to play every match in hardcore carry mode because they've got a bunch of dead weight stacked on them.

So we need to go back to a good matchmaker. That matchmaker needs built around w/l, like every other good matchmaker in the world. However people making stupid arguments like 'w/l isn't accurate for QP' make it hard to even start that discussion. More to the point there's a toxic mentality, one of 'it's not me, it's my team in QP'. I stomp on that attitude when it comes up because it's false and like most terribad ideas designed as an excuse for why someone is not succeeding it spreads like poison in the well.

Everyones win/loss is a direct reflection of how well you help your team win matches. It's accurate and if someone wants to win more they need to take better mechs and learn better habits. It's absolutely not random - it's not possible to be random, because you're in every match you play in. As such your presence influences the match.

You're always observing in your own match. There's no question if the cat is alive or dead, the box never even closed. You're there, influencing which way the particle goes.

[Xavori]

MischiefSC, on 28 November 2017 - 01:13 PM, said:

You also seem to absolutely and totally not understand how matchmakers and win/loss work. Neither Elo nor TrueSkill push a player to 1.0 w/l. That's why you have a matchmaker with a different ranking instead of just using win/loss.

It assigns you a value and then modifies that value based on who you played against - so beating a player who's better than you raises you a lot, someone who's your skill a tiny bit, someone less than you not at all. So your win/loss only every changes by 1 but your ranking changes by a variable number based on your team and the other team.

That bit right there is where you are wrong, and everything that follows in your post is therefore flawed.

ELO ranking is specifically meant that when two players/teams of identical ranking play each other, they will have an equal number of wins and losses. In other words, if they have the ranking, they will both have a W/L ratio of 1 vs each other. And then you add in a matchmaker that is attempting to always get equally skilled players on both sides, which means all players will eventually have a W/L ratio of ~1.

That's fundamental to ELO ranking. Like, it's the seriously most important thing Elo had in mind when coming up with it because he used a win probability of 50% (ie. a W/L ratio of 1) as 'proof' that the two players had the same skill.

It's also fundamental to TrueSkill and virtually every other quality player ranking system in use. If two equally skilled opponents/teams compete against each other, each of them will win 50% of the time. This means that W/L WILL NEVER BE A MEANINGFUL STAT IN A TRUE RANKING SYSTEM.

Your failure to understand that is what's driving part of your problem with my statements about the leaderboard. We don't have a matchmaker that does that which means we don't have a matchmaker that produces meaningful results. Random numbers are random.

As for the nonsensical human ability curve...you can't quantify it. You can't set a range. Therefore, you have no expected value no matter how large a sample size you take. So no, you can't try to use any numerical analysis to come up with something meaningful. You'd just be sticking random numbers in to the formulas, and random numbers are random.

I know you want to believe that your above 1 W/L ratio means something. It does. It means that FOR WHATEVER REASON you've ended up on teams that were more often better than the other team. Now, you might be actually be an above average skill player. You might just be getting really lucky. You might be logging in at a time when severely potato players are in the majority. You have no idea why your W/L ratio is above 1 that you can claim with any certainty tho which has been my point this whole time.

What you can know with certainty tho, is that because most of the W/L ratios in the leaderboard are not ~1, the matchmaker is doing a crap job of creating quality matches. Then again, we don't need the leaderboard to know that. We've all seen just how many 12-3 or worse stomps happen.

[Xavori]

MischiefSC, on 28 November 2017 - 02:12 PM, said:

That didn't work, all it did was add mediocres and bads into gameplay with good players, resulting in the bads getting farmed. Most matches in T1 are more like an 8 v 8 or 6 v 6 with farmables filling in. This is frustrating for everyone. The bads get thrown up to T1 where they settle into a sub 1.0 w/l and get destroyed match after match after match, rarely getting kills or decent damage. The good players have to play every match in hardcore carry mode because they've got a bunch of dead weight stacked on them.

I actually agree with everything there except the last sentence cuz I don't do that. Ya, sometimes I'll be feeling froggy, I'll jump in a comp mech and play for really reals and rack up kills and damage, but other times, like the last two days, I'll be trying to make bad things work.

For example, my light gauss built Ghille with full jumpjets, max engine size, and sadly, stripped down to light level armor to make the rest of this fit. I actually got this thing to where I'd consistently do 400-700 damage. But the problem was that in doing so, I was leaving my team a mech down for the first ~5 minutes of every match while I got into good position to rapid fire light gauss rounds since a big part of what makes light gauss work is a dps advantage over regular gauss (also a small potential damage advantage in the ammo per ton,). Needless to say, a 12v11 match usually ends up 12v10 then 12v9 in very short order. And by the time I'd be in position and start my pouring in light gauss rounds, the match pretty much lost.

But I gave it ~30 quick play rounds making small tweaks the whole time to see if I could get it to work. Sadly, it doesn't and instead I've had to go back to the same regular gauss and cut some of the jump jets and whatnot and just fight with my team instead of getting optimal sniper positions.

Anyway, the point is not every good player goes into hard carry mode every match cuz holy hell would that get boring in a hurry.

[Asym]

Guys, give it up...... It doesn't matter who is right, really!

PGI is not going to change anything, ever......

A neat discussion but nothing can or will change.

It's a draw: like the past events.......no winner.......no losers.

[MischiefSC]

Xavori, on 28 November 2017 - 02:20 PM, said:

That bit right there is where you are wrong, and everything that follows in your post is therefore flawed.

ELO ranking is specifically meant that when two players/teams of identical ranking play each other, they will have an equal number of wins and losses. In other words, if they have the ranking, they will both have a W/L ratio of 1 vs each other. And then you add in a matchmaker that is attempting to always get equally skilled players on both sides, which means all players will eventually have a W/L ratio of ~1.

That's fundamental to ELO ranking. Like, it's the seriously most important thing Elo had in mind when coming up with it because he used a win probability of 50% (ie. a W/L ratio of 1) as 'proof' that the two players had the same skill.

It's also fundamental to TrueSkill and virtually every other quality player ranking system in use. If two equally skilled opponents/teams compete against each other, each of them will win 50% of the time. This means that W/L WILL NEVER BE A MEANINGFUL STAT IN A TRUE RANKING SYSTEM.

Your failure to understand that is what's driving part of your problem with my statements about the leaderboard. We don't have a matchmaker that does that which means we don't have a matchmaker that produces meaningful results. Random numbers are random.

As for the nonsensical human ability curve...you can't quantify it. You can't set a range. Therefore, you have no expected value no matter how large a sample size you take. So no, you can't try to use any numerical analysis to come up with something meaningful. You'd just be sticking random numbers in to the formulas, and random numbers are random.

I know you want to believe that your above 1 W/L ratio means something. It does. It means that FOR WHATEVER REASON you've ended up on teams that were more often better than the other team. Now, you might be actually be an above average skill player. You might just be getting really lucky. You might be logging in at a time when severely potato players are in the majority. You have no idea why your W/L ratio is above 1 that you can claim with any certainty tho which has been my point this whole time.

What you can know with certainty tho, is that because most of the W/L ratios in the leaderboard are not ~1, the matchmaker is doing a crap job of creating quality matches. Then again, we don't need the leaderboard to know that. We've all seen just how many 12-3 or worse stomps happen.

Again. I realize that you haven't read anything on any of this and keep treating your opinions like facts, but Elo is a name. It's not all capitalized.

All human skill is on a curve. It's a finite range. You absolutely can quantify skill. It has limits.

Infinity

There's another link for you. Please tell me how human skill is infinite and, as such, unquantifiable. Every major business in the world quantifies, measures and values out human skill. It's called human capital. Every profession in the world measures skill. Those who measure it for anything like a comparative or matchmaking purpose do so on a curve.

Again, back to your TrueSkill link, please look at pages 5-15. It goes into detail exactly how and why they use a Gaussian curve for skill distribution instead of a Logistic one.

You're literally talking about TrueSkill, which is 100% based off of win/loss, and then saying that no meaningful ranking system uses win/loss. That's all they lose. Who you won or lost against, what its expectation for you was and what its confidence in that expectation was given the match that was created.

Also humans are not static. Two equally ranked players will NOT keep a 50% win/loss. There will be a bit of variance because they will be learning every single match and at different rates and different skills. There's always variation, it's just within a given range. The ranges you use to break players up is called a deviation. Again, this is covered in the TrueSkill system you linked to me. Here's the link again so you don't have to find it.

Read it. Pages 5-15 at least.

[MischiefSC]

Asym, on 28 November 2017 - 02:32 PM, said:

Guys, give it up...... It doesn't matter who is right, really!

PGI is not going to change anything, ever......

A neat discussion but nothing can or will change.

It's a draw: like the past events.......no winner.......no losers.

Nice try. I appreciate the peacemaking.

However it's not a draw. PGI does need a new matchmaker and it needs to be w/l based like every successful matchmaker in games. Ideally it's got a score for players, a score for mechs and a modifier based on loadout but we may not even need that degree of accuracy. It can only build matches out of who's available and match for tonnage.

The delusion that players are utterly helpless in the face of an utterly random team is demonstratively false. By trying to give it a sense of legitimacy and say it's a tie is to say that a blatant, self serving lie is true.

I'm not a fan of that.

If someones win/loss is lower than they want it's not because their teams are bad, it's because they are not as good as they want to be. Everyone is playing in the same mix of teams. If someone plays better, they win more. You're 8.333% of your team. How much that 8.333% can carry is all you.

[Xavori]

MischiefSC, on 28 November 2017 - 02:38 PM, said:

If someones win/loss is lower than they want it's not because their teams are bad, it's because they are not as good as they want to be. Everyone is playing in the same mix of teams. If someone plays better, they win more. You're 8.333% of your team. How much that 8.333% can carry is all you.

You cannot prove that anymore than I could prove that the reason a person has an above average W/L ratio is because they only log in when nothing but potatoes are playing. You simply are making an assumption you want to be true.

I'm not saying it's not true that better players have an above 1 W/L ratio in the current environment. I'm saying neither I nor anyone else can say with any certainty that that's the case because the matchmaker is random.

What I can say is that because W/L isn't ~1 for most of the playerbase, the leaderboard is garbage. But I was saying that before because I already know it's using basically random teams and random teams produce random results and all together now...random numbers are random.

p.s. I capitalize ELO the skill system to differentiate from Elo the person.

p.p.s. If you want an easy to read explanation of implementing ELO to non-chess, 538 came up with an ELO-based system for the NFL. The thing that should jump out at you as you read it...it's not just wins and losses.

https://fivethirtyei...fl-elo-ratings/

[MischiefSC]

You don't capitalize the name of the system. We're not discussing people, there's nothing here to make anyone confused about the Elo ranking system vs Arpad Elo the person.

As to 538, the system is based off win/loss. The score is win/loss based. It functions exactly like Elo, save that it adds an additional modifier based on margin of victory. So a big win moves your score up more than a close win. It doesn't take into account individual player stats, aside from wins and losses.

You're literally pointing out, again, that you don't understand the math involved. Like Elo, every team has a base score of 1500. That score is adjusted up or down based on who they win or lose against. If they beat a better team it goes up by more than if they beat a bad team. He just adds the qualifier that he adjusts the score a bit more based on big wins or losses.

However, all that matters in that system is win/loss. It doesn't take into account who throws the most passes or rushes or if they use a running game or a passing game. It doesn't track interceptions or anything else.

It counts win or loss. Then adjusts both scores. Because he wants a wider distribution (too few teams, you'd end up with many teams in a very, very narrow band otherwise) he makes an additional adjustment up or down based on margin of victory.

Beside which NFL, which plays regular teams, is nothing like QP, which is only your own stats. However it is relevant to prove that win/loss is relevant for consistent teams as well.

You don't seem to understand what random is either. If win/loss was random then your win/loss would be, and I'll quote the dictionary here, "of or characterizing a process of selection in which each item of a set has an equal probability of being chosen". That means you're as likely to have a 10.0 win/loss as a 0.01. That's random. That would be a random result.

Your last 3 months of stats show you with

season 17 at 49.4% wins
season 16 at 46.4% wins
season 15 at 48.6% wins.

Asym, for another reference, is


season 17 at 45.6% wins
season 16 at 43.3% wins
season 15 at 45.6% wins.

Season 17, the name right above Xavori on the leaderboard is GeminiWolf. No idea who they are, don't care.

season 17 at 51.0% wins
season 16 at 51.0% wins
season 15 at 48.1% wins.

Season 17, the highest win/loss on the same page as Xavori, is DAEDALOS513, who I see a lot in QP.

season 17 at 68.4% wins
season 16 at 61.3% wins
season 15 at 61% wins.

For sake of fairness, here are my own stats. I have never played 10 matches in group queue in a season in, gah, years.

season 17 at 68.5% wins
season 16 at 65.0% wins
season 15 at 61.7% wins. *caveat - only 47 matches in QP this season*

I could go on and on but this is a good example. These are not random. Nor would it be, humans function within set ranges.

So, again, here's mathematical proof you are wrong.

[Xavori]

MischiefSC, on 28 November 2017 - 03:22 PM, said:

You don't capitalize the name of the system. We're not discussing people, there's nothing here to make anyone confused about the Elo ranking system vs Arpad Elo the person.

As to 538, the system is based off win/loss. The score is win/loss based. It functions exactly like Elo, save that it adds an additional modifier based on margin of victory. So a big win moves your score up more than a close win. It doesn't take into account individual player stats, aside from wins and losses.

You're literally pointing out, again, that you don't understand the math involved. Like Elo, every team has a base score of 1500. That score is adjusted up or down based on who they win or lose against. If they beat a better team it goes up by more than if they beat a bad team. He just adds the qualifier that he adjusts the score a bit more based on big wins or losses.

No. Right there in that paragraph you're pointing out exactly what I've been saying. IT'S NOT JUST WIN LOSS. Quality matters. That's why beating a good team is worth more than beating a bad team.

What's more, you also finally acknowledged that you're not looking at W/L ratios when looking at ELO (Yes, I can totally keep capitalizing if I was so ppppbbbbttttt!!!). You're looking at a rating based on your W/L vs the quality of your opponents and any other considerations built in to the equation (like with TrueSkill which builds in piles of them). Winning and losing is only used to determine direction of movement. Literally everything else involved in the rating is something other than winning or losing.

As for trying to pull meaning out of my stats on the leaderboard... *falls down laughing and dies*

If I'm so inclined, I can do this:
https://steamuserima...3A1DB72472978D/
(ie. KMDD almost half the other team)
or in faction do this:
https://steamuserima...A0800697B3A87E/
(ie. basically KMDD an entire wave worth of mechs +1...and doing it using nothing heavier than 65 tons just to see if it was possible)

and then I'll turn around and do something like this while trying to make a new (and silly) build work like a light gauss high speed Ghille:
https://steamuserima...3749171F819DE5/
(so painful)

until I actually did get the little tweaks figured out and then how to pilot it so it all came together:
https://steamuserima...250B80685B3AA5/
(sadly, not enough to carry the team tho, and it highlighted the problem with not being able to secure my own kills before my team steals them when using a light gauss at range *smirk*)

Oh, and if you're feeling froggy enough to want to see if you can get a light gauss Ghillie above 400 or so damage, here's the build:
https://steamuserima...6163D644890A4A/
I wish you luck. It's badly undergunned and as delicate as a rose petal in a meat grinder.