76 replies to this topic

[PhoenixFire55]

Deathlike, on 17 April 2017 - 01:12 PM, said:

You still already have flaws in your implementation. I'll try to break it down in other sections.

Looks like another one who didn't read anything of what I've said. My "implementation" of what? ... I never suggested any implementation of anything since I've never suggested any criteria that would be used to evaluate skill or anything alike.

Deathlike, on 17 April 2017 - 01:12 PM, said:

You're making a few assertions though. It's not like it's one player in like CS:GO that has access to whatever. You have to have AT LEAST picked a mech. Someone picking a Mist Lynx is far different from someone else picking a Kodiak-3. It's like trying to put Proton in the worst mech possible vs some T5 unicorn in a Kodiak-3. They are drastically different aspects that you're just holding onto just 1 number for. It doesn't work.

Amazing. Include these and whatever else factors into the criteria MM uses for evaluating peoples performance if you like. Everything that I've said remains unchanged. You still need 24 people.

Deathlike, on 17 April 2017 - 01:12 PM, said:

So in your MM example, what if the result is having 12 Lights on one side and 12 Assaults on the other? If we go STRICTLY by the numbers, such a result can happen. While obviously my example is an extreme one, but that already happens in the solo queue as is. What you forget or mistake is that you assume that the distribution of mechs/tonnage is relatively even... which still requires more people/population and you can't magically lump them all in the same boat. It just makes the MM result much worse and broken.

I never provided an example of a working MM. I've provided an example how 24 people is all that it takes for a perfectly balanced match as long as their "skill" or impact on a match is measured correctly. I've never provided any solutions on how to measure that skill, because it was not what I was discussing in the first place.

If people and their mechs impact are evaluated properly then it is irrelevant what mechs they are in. Matching classes is a bandaid. A bandaid that arguably takes away more than it achieves.

Deathlike, on 17 April 2017 - 01:12 PM, said:

You can't really excuse it either. It's like saying, everything is broken and we should keep it broken because it's fine. Yea, that'll go far.... unless you want to be taken seriously.

Excuse what? ... My only intention was exactly to show how blaming lack of population is nothing but an excuse for their lack of ability, effort or whatever to build a proper MM. You want me to create a proper MM instead of PGI? ... No thanks, I'm not getting payed for this.

Deathlike, on 17 April 2017 - 01:12 PM, said:

It's not just 24 people, it's 24 mechs and not everyone is necessarily as effective in every mech (remember, there will be people grinding in mechs they are unfamiliar with or in a class they are not as good at). Simplifying and stripping relevant criteria is only going to make the MM result worse... unless you don't care about the result (which is already what PGI is doing anyways).

Read above. I've already covered that.

Deathlike, on 17 April 2017 - 01:12 PM, said:

There's a reason why filters exist (like even the region filters) affects how effective an MM is. If only a 1/3 of the population actively runs Oceanic (which doesn't ever happen for me, since I totally deselect it), it affects the quality of your results much more significantly... and it would still be the case for tonnage/weight classes in these examples. It's not even that hard to find 6 Assaults on one side vs 0 to 2 Assaults on the other side in a normal group queue match. These are things that will get people whining about stuff that much more readily.

And for heck knows what time again ... that all is irrelevant. You need at least 24 people for a match, and 24 people of any kind in any mechs is more than enough for a perfectly balanced match if done right.

BLOOD WOLF, on 17 April 2017 - 01:16 PM, said:

It also looks like you took my quote out of context. I was saying "you don't understand" to you, not jeffers. You made it look like I was saying that to mr jeffers. Or am I reading that wrong?

I've quoted the entire post, the don't understand part is irrelevant, the one where you agree with him talking about "multi-dimensional arrays" tho is since you deny that you are the one using such terminology.

[MrJeffers]

PhoenixFire55, on 17 April 2017 - 12:47 PM, said:

LOL, that's the best you could come up with? Once again clearly showing that you don't understand a thing ... Missing the one simple point over and over and over again.

It doesn't matter who the players in question are, doesn't matter from what "ranks" or "tiers" they are, doesn't matter what mechs they bring and what loadouts they use and whatever else you can think of. Any player in any mech has a limited amount of impact he can have on a match. That amount is determined by his skill, by his mech, by his loadout, by whatever. The important thing is that it all translates into one thing, one amount. Properly evaluating that amount is not a trivial task, but never did I claim that I provided or even mentioned anything about how it should be done. That is a task that any developer creating a MM system solves, or in our case fails to. That is why matches are unbalanced. Because as long as there is a proper way of evaluating the impact a player makes, the amount of players becomes irrelevant and all it takes for a balanced 12v12 match is 24 people of any skill, in any mechs and whatever else.

In your silly example with mech classes you can simply assign a multiplier for each mech class, or each mech chassis, or each mech variant, or even each mech loadout that together with the "skill" level will combine into "magic value no.33" that the MM will in turn use. These multipliers can be even calculated by the MM itself based on average performance of said mechs, their variants or loadouts, or individual performance of any particular player in said mechs, variants and loadouts. It takes nanoseconds for modern PCs to perform such operations. But once again, this isn't what I am talking about here at all.

My "silly example" is exactly what the match maker is dealing with. Go ahead and apply your mech class multiplier and randomly apply them across your 24 number in the approximate ratios of the queues. Then try and build your match. It's not going to work, and it's not going to be balanced. Even making the massive assumption that both the skill category AND the mech multiplier are accurate. Good luck with that. If it was that simple it would have been done already.

Edited by MrJeffers, 17 April 2017 - 01:33 PM.

[BLOOD WOLF]

PhoenixFire55, on 17 April 2017 - 01:27 PM, said:

can we save ourselves the trouble. Dude your being to simplistic with just 24. We already know the MM accounts more more than just 24 numbers.

"24 people of any kind in any mehcs is more than enough for a perfectly balanced match if done right". Nobody is arguing with you there. you are correct; However given how it works, that scenario is rarely ever going to play out. If the probability was calculate the number of that would be very low. I bet 1.X% if lucky

Edited by BLOOD WOLF, 17 April 2017 - 01:33 PM.

[PhoenixFire55]

MischiefSC, on 17 April 2017 - 01:16 PM, said:

To clarify - suppose you're on an assembly line getting random fruit. You're trying to put them in packages with everything in pairs - 2 apples, 2 oranges, etc. Chute opens and..... you get 7 potatoes, 1 kiwi, 3 oranges, 5 assorted apples (different types), a banana, a cherry, 1 small and 1 large wedges of cheese, and 3 flavored condoms.

Now put that into 2 equal, balanced sets.

Easy. Formulate a balancing criteria, i.e. how many oranges balances one apple, etc.

And if you are going to tell me that there is no such criteria and oranges and apples are fundamentally different then your example is fundamentally different from a MM. Because if you deem all people different, then there is no perfect match for any of us, thus balanced matches are impossible. But since we all regularly see examples of the opposite (i.e. balanced matches do happen), it means that people are not different, but rather balancable.

[BLOOD WOLF]

MrJeffers, on 17 April 2017 - 01:32 PM, said:

If it was that simple it would have been done already.

His problem is he is failing to understand a flaw in his logic, or his explanation is blaming PGI and they just need the perfect MM which nobody has ever done before.

The only perfect MM is a theoretical one, not a practical one

[PhoenixFire55]

MrJeffers, on 17 April 2017 - 01:32 PM, said:

My "silly example" is exactly what the match maker is dealing with. Go ahead and apply your mech class multiplier and randomly apply them across your 24 number in the approximate ratios of the queues. Then try and build your match. It's not going to work, and it's not going to be balanced. Even making the massive assumption that both the skill category AND the mech multiplier are accurate. Good luck with that. If it was that simple it would have been done already.

LOL ... I've already showed how 24 completely random players can be perfectly balanced. So if I apply these "random" multipliers to a random set of 24 values I already have I will get ... *drumroll* ... a different set of 24 random values. I.e. nothing different for the purpose of balancing.

As for being "accurate", never did I suggest anything about how such criteria should be evaluated. I've said that like a dozen times already, but since you don't bother reading ...

[BLOOD WOLF]

PhoenixFire55, on 17 April 2017 - 01:39 PM, said:

LOL ... I've already showed how 24 completely random players can be perfectly balanced. So if I apply these "random" multipliers to a random set of 24 values I already have I will get ... *drumroll* ... a different set of 24 random values. I.e. nothing different for the purpose of balancing.

As for being "accurate", never did I suggest anything about how such criteria should be evaluated. I've said that like a dozen times already, but since you don't bother reading ...

pheonix, if you create a theoretical situation of course you can make it perfect. I have 24 people with X attributes and for 1 there is another that can be perfectly matched. Sure, but that's not practical.

how many teir 1's and 2's seen a trial mech or person play so badly just this month

Edited by BLOOD WOLF, 17 April 2017 - 01:41 PM.

[PhoenixFire55]

BLOOD WOLF, on 17 April 2017 - 01:33 PM, said:

"24 people of any kind in any mehcs is more than enough for a perfectly balanced match if done right". Nobody is arguing with you there.

Yes they do, plenty of people actually.

BLOOD WOLF, on 17 April 2017 - 01:33 PM, said:

you are correct; However given how it works, that scenario is rarely ever going to play out. If the probability was calculate the number of that would be very low. I bet 1.X% if lucky

Maybe. I don't really expect them to figure out how to properly estimate "skill" and other things affecting matches. However, once again, that has nothing to do with the point I am making, i.e. the fact that MM fails have absolutely nothing to do with player population.

[BLOOD WOLF]

PhoenixFire55, on 17 April 2017 - 01:41 PM, said:

I don't really expect them to figure out how to properly estimate "skill" and other things affecting matches. However, once again, that has nothing to do with the point I am making, i.e. the fact that MM fails have absolutely nothing to do with player population.

Okay, that is where I agreed with you. The problem is the followup to that conclusion that since it has nothing to do with pop, then it's obviously PGI's lack of coding skills.

Which leads back to the topic, and No, its not. It happens to be because of various sets of variables that nobody can really control. Well actually they can, but then we get a whole other set of problems.

maybe the MM should take a minute max to build a match before launching.

Edited by BLOOD WOLF, 17 April 2017 - 01:50 PM.

[Deathlike]

PhoenixFire55, on 17 April 2017 - 01:30 PM, said:

Looks like another one who didn't read anything of what I've said. My "implementation" of what? ... I never suggested any implementation of anything since I've never suggested any criteria that would be used to evaluate skill or anything alike.

I've read what you said, and what you said ignores everything that matters when it comes to actual matchmaking. You're treating it as a singular problem when it is in fact a multifaceted one.

Quote

Amazing. Include these and whatever else factors into the criteria MM uses for evaluating peoples performance if you like. Everything that I've said remains unchanged. You still need 24 people.

You can't, because you need more than 24 people to evaluate from. You'll end up figuring out 24 people from at least hundreds and thousands of them. While 24 will play the match, you still have to get them from a pool of players.

Quote

I never provided an example of a working MM. I've provided an example how 24 people is all that it takes for a perfectly balanced match as long as their "skill" or impact on a match is measured correctly. I've never provided any solutions on how to measure that skill, because it was not what I was discussing in the first place.

If people and their mechs impact are evaluated properly then it is irrelevant what mechs they are in. Matching classes is a bandaid. A bandaid that arguably takes away more than it achieves.

It's partly a bandaid, but there's a reality. If you like mismatched number of Assaults or Lights (particularly the latter for Domination maps like Crimson Strait that requires one to even get to the circle on a particular side), people are not going to have fun when it ends before it begins (and I've seen those kinds of matches before, unfortunately).

You're also putting heavily on the way the MM rates players... which even if perfect, is possible to construct poor teams (or generating lopsided results for that matter). It's like putting a comp team together with great players, but having not played with each other a lot and expect to win Div-A MRBC. It doesn't work.

Quote

Excuse what? ... My only intention was exactly to show how blaming lack of population is nothing but an excuse for their lack of ability, effort or whatever to build a proper MM. You want me to create a proper MM instead of PGI? ... No thanks, I'm not getting payed for this.

Lack of ability is certainly a thing I criticize. However, whatever your complaints/suggestions are only going to amplify the problem, not solve it. You're simplifying the problem ignoring important facets that are legitimate issues with the system.

Quote

And for heck knows what time again ... that all is irrelevant. You need at least 24 people for a match, and 24 people of any kind in any mechs is more than enough for a perfectly balanced match if done right.

You're making assertions that don't necessarily create/predict/construct the scenario you wish to happen. It's not even a "more likely than not" as it is closer to "a snowball's chance in hell".

Edited by Deathlike, 17 April 2017 - 01:56 PM.

[PhoenixFire55]

Deathlike, on 17 April 2017 - 01:52 PM, said:

I've read what you said, and what you said ignores everything that matters when it comes to actual matchmaking. You're treating it as a singular problem when it is in fact a multifaceted one.

I never intended to solve a MM problem here. As I've clearly stated multiple times. But since you've read what I said ...

Deathlike, on 17 April 2017 - 01:52 PM, said:

You can't, because you need more than 24 people to evaluate from. You'll end up figuring out 24 people from at least hundreds and thousands of them. While 24 will play the match, you still have to get them from a pool of players.

No. If I somehow know the evaluation method I don't need anyone at all to be able to evaluate properly. And if I do not, then all I need is 24 people playing against each other in various combinations to gather stats and iteratively reach the close enough approximation. The more matches, the closer such approximation will be to the ideal evaluation. The size of the pool I need is exactly the minimal amount of people it takes to form a match.

Deathlike, on 17 April 2017 - 01:52 PM, said:

It's partly a bandaid, but there's a reality. If you like mismatched number of Assaults or Lights (particularly the latter for Domination maps like Crimson Strait that requires one to even get to the circle on a particular side), people are not going to have fun when it ends before it begins (and I've seen those kinds of matches before, unfortunately).

The ability of a light to cap, control the circle or whatever else should obviously be reflected in what I've said is "impacting" the match. Same way for the ability of an assault to do their thing. Also, there is nothing saying that different mechs can't have different balancing multipliers for different game modes, same way they can have different multipliers for different people driving them. That is obviously also a part of proper criteria.

Deathlike, on 17 April 2017 - 01:52 PM, said:

You're also putting heavily on the way the MM rates players... which even if perfect, is possible to construct poor teams (or generating lopsided results for that matter). It's like putting a comp team together with great players, but having not played with each other a lot and expect to win Div-A MRBC. It doesn't work.

Player-with-player synergy can also be accounted for. Ain't saying its easy, but it is by all means possible.

Deathlike, on 17 April 2017 - 01:52 PM, said:

Lack of ability is certainly a thing I criticize. However, whatever your complaints/suggestions are only going to amplify the problem, not solve it. You're simplifying the problem ignoring important facets that are legitimate issues with the system.

I do not complain and I do not suggest anything.

Deathlike, on 17 April 2017 - 01:52 PM, said:

You're making assertions that don't necessarily create/predict/construct the scenario you wish to happen. It's not even a "more likely than not" as it is closer to "a snowball's chance in hell".

Maybe, maybe not. I do not evaluate the likelihood of PGI coming up with a proper balancing criteria either.

[Sanlucif3r]

what is up with the match maker!?!

why are there so many lrm atlas in my teams!?????

why do light mechs run straight into the enemy and suicide at minute 1 of the match??

have we lost so many players that this is all we can get now???!?!?!?!?!

[Dino Might]

Some are missing that we aren't making one match with 24 players from a pool of 240. We are trying to make anywhere between 5 and 10 matches concurrently from that pool. That changes the goal in how to balance each match. Think least squares or another curve fitting technique on your metrics. The best answer may be that none of the matches are "perfectly" balanced.

[MischiefSC]

PhoenixFire55, on 17 April 2017 - 01:35 PM, said:

Easy. Formulate a balancing criteria, i.e. how many oranges balances one apple, etc.

And if you are going to tell me that there is no such criteria and oranges and apples are fundamentally different then your example is fundamentally different from a MM. Because if you deem all people different, then there is no perfect match for any of us, thus balanced matches are impossible. But since we all regularly see examples of the opposite (i.e. balanced matches do happen), it means that people are not different, but rather balancable.

You'd need near perfect prediction to make it work with a pool of 24 players. What you're asking for is a better predictive analysis with 24 randoms than bookies have for static professional sports teams.

It's just not realistic. You're asking for a better matchmaker than TrueSkill can represent, which is probably the single most expensive matchmaking system (well, ranking system to support a mm) ever produced.

It's trying to balance the best possible option among numerous simultaneous matchups while matching both for skill variance (in a general sense) and tonnage variance (in a general sense). Even if it had a sliding Elo score for each player based on specific mechs, variants and loadouts (I'm better in a 12 ersml Nova than an SPL one for example) it would only mildly impact the actual breakdown of teams because it's still got to pick from the people available in that 30/60/90 second increment and tonnage.

[PlayerUnknown]

You're mad because you're not on The Winning Side that often

So hit me up I will teach you the ways of the Er p PC

And yell at you constantly until you get it it will be great

[PhoenixFire55]

MischiefSC, on 17 April 2017 - 04:35 PM, said:

You'd need near perfect prediction to make it work with a pool of 24 players.

For the first try sure, but as the game does gather stats the more games each person plays the closer the estimation of his skill would be to the actual value. Of course as long as they are playing in different combinations.

MischiefSC, on 17 April 2017 - 04:35 PM, said:

It's just not realistic. You're asking for a better matchmaker than TrueSkill can represent, which is probably the single most expensive matchmaking system (well, ranking system to support a mm) ever produced.

I am not asking for anything. I'm stating a simple fact that MM fails have nothing to do with player population.

MischiefSC, on 17 April 2017 - 04:35 PM, said:

It's trying to balance the best possible option among numerous simultaneous matchups while matching both for skill variance (in a general sense) and tonnage variance (in a general sense). Even if it had a sliding Elo score for each player based on specific mechs, variants and loadouts (I'm better in a 12 ersml Nova than an SPL one for example) it would only mildly impact the actual breakdown of teams because it's still got to pick from the people available in that 30/60/90 second increment and tonnage.

And once again, that is all irrelevant. Since skill, mechs, loadouts and whatnot combines into one value, all you need is enough players to form a match, i.e. 24. As long as you have 24 players in whatever timespan and as long as your estimation of said factors is correct you will have a perfectly balanced match each time. The estimation is the difficult part, however I've never said I have an answer or even aim to discuss it here. However it is exactly this estimation that causes MM imbalance, not anything else.