182 replies to this topic

[Quaamik]

I think I people are misunderstanding the MM.

It cannot give you a even chance of winning / losing. It definately cannot ensure an even (or close) number of wins to losses. It can only try to ensure your matched against people with equal skill, which it supposedly does by matching your win / loss ratio with others. Most people are not going to complain if they lose if the reason is they sucked that game, or their team didn't jell. Where they are upset is when it's obvious they were outclassed from the start, or their teammates were obviously not of their skill level.

Maybe that itself is the problem.

Take this example. Player A plays with his buddies mostly. He's in a high end competitive group. They either play against themselves in matches for practice, or drop against other competitive groups. Occasionally they wind up in a match with average folk. Because of always playing against people with his skill level, his ELO is around 55%.

Player B loaded the game 2 months ago. He's got some skill, and is mainly playing against other new players. As he improved, his skill level increased a bit, and because most of the players on when he is are around the same skill level, he's also at an ELO of 55%.

The difference is that Player A has played 500 matches. Player B has played 50.

Does the matchmaker drop them both in the same match when player A drops in a PuG match? Can it tell that player A has a 80% hit average (overall) while player B has a 25% hit average?

Maybe it should be looking at something other than ELO.

[topgun505]

Quaamik, on 06 December 2014 - 07:51 PM, said:

I think I people are misunderstanding the MM.

It cannot give you a even chance of winning / losing. It definately cannot ensure an even (or close) number of wins to losses. It can only try to ensure your matched against people with equal skill,

*snip*

Equal skill? AYFKM? Game after game after game my teams seem to get rolled and half of my team (especially the assault pilots) and pulling in damages of 100-150 whereas the other side is pulling 500, 600, 700. I'm literally getting 2-4x the damage of the assaults in my Locust. Let that sink in a minute. A Locust is pumping out 4x more damage than those people. Once in a while with that happening ... sure ... no problem. Understandable. But game after game (like 20 in a row). I REALLY have a hard time with that above statement. It most definitely does NOT pit those of equal skill against each other.

[Abivard]

Sandpit, on 06 December 2014 - 04:20 PM, said:

And ignorant uninformed opinions abound.

Read the factual data some time. It's very enlightening as opposed to spouting ridiculous unfounded conspiracy theories

Just what unfounded theories am I spouting?
That for every truth you speak you feel compelled to add a falsehood?

[Duke Nedo]

Sandpit, on 06 December 2014 - 04:01 PM, said:

The mm doesnt create anything. That's where the fallacy comes in and what those who understand the mm are trying to explain

The mm doesn't punish
It doesn't reward
It doesn't do anything except match you to players in an elo range. The mathematical equations used to do that don't change no matter how many times in a row you win or lose.

Your statistical analysis has no basis in regards to mm. You have a 50/50 shot of winning each and every game. There's no % involved in thr mm system regarding your w/l for anything other than calculating your elo after a match.

That's it. Nothing more. nothing less.

Ok mister, you really tick me off now. You blurt out "factual truths", when in fact you and me have access to exactly the same information. If you had been the coder of the MM you could make statements like that. Fact is, you don't know and I don't know anything more than what PGI has told us.. and that leaves out a lot of critical information. As long as we cannot use the information they have shared to fully reproduce the MM ourselves, they have not described how it works.

That does not make you enlightened or me uninformed. It just makes you arrogant and ignorant.

[Abivard]

topgun505, on 06 December 2014 - 09:48 PM, said:

*snip*

Equal skill? AYFKM? Game after game after game my teams seem to get rolled and half of my team (especially the assault pilots) and pulling in damages of 100-150 whereas the other side is pulling 500, 600, 700. I'm literally getting 2-4x the damage of the assaults in my Locust. Let that sink in a minute. A Locust is pumping out 4x more damage than those people. Once in a while with that happening ... sure ... no problem. Understandable. But game after game (like 20 in a row). I REALLY have a hard time with that above statement. It most definitely does NOT pit those of equal skill against each other.

Try reading the rest of his post... MM tries to match players by skill, however.... what it uses to try and measure player skill is broken and indeed doesn't measure player's skill at all.

Which results in en example of the old adage:

"garbage in, garbage out." Which is why MM SEEMS to be failing, MM isn't, what is failing is the assignment of skill values to players.

Edited by Abivard, 06 December 2014 - 11:16 PM.

[Sandpit]

occusoj, on 06 December 2014 - 06:16 PM, said:

In fact it isnt their personal skill in some matches.

???
in the exact statement you quoted I specifically said
"OR THEIR TEAM PLAYING POORLY"
smh

occusoj, on 06 December 2014 - 06:16 PM, said:

In fact it isnt their personal skill in some matches.
As stated, Ive not fired a single shot in some stomps for the sake of testing the impact. My team rolled over the enemy 12-0 or at worst 12-2. Looking at the stats its obvious that 3 or 4 players dealt serious damage, the rest is much below that. One individual player on the loosing team couldnt have prevented that as engagements were always 4-10 or worse for them.
The individual has extremely little input only in those matches that already put the loosing team behind a couple of mechs because they either have some AFK, extreme derpups or half of their assaults are LRM atlai.

Team performance is a big part of it, if they spread out, suicide, stop in chockepoints,.... and the other team doesnt, they will get stomped. As an individual its still frustrating if that happens like 10 times in a row.

And as I stated in my first post, MM has currently no way of knowing enought about the players to prevent total stomps.

I dont blame MM for my poor skill, I know Im a bad player all by myself. Still wondering why I often dont get matched against similiar bads.


Well, thats exactly what you actually state.
Theres a 50/50 chance with each match to win or loose. Its like flipping a coin.
Each flip has an individual chance of 0.5 to yield result A (and 0.5 for .
A streak is nothing more than a series of flips that each have the same result.
So flip the coin, get side A. Flip again, side A. AND side A, AND side A, AND side A,....
Each individual flip still has the same 0.5 chance of resulting in A and B. For the whole series of flips to have the same result its 0.5^n.

Or firing an UAC. Each individual (double tapped) shot has a given chance of jamming the weapon. Theoretically you could fire 100 such shots without a jam, its not impossible but extremely rare.

???
Is there some disconnect I'm missing here? You just agreed 100% with what I said which is in direct contradiction of the player's post I was responding too that implied the MM had some sort of mechanic and coding that sets up players to have streaks in that their w/l ratio is determined by the MM system...

Quaamik, on 06 December 2014 - 07:51 PM, said:

I think I people are misunderstanding the MM.

trust me, there's not think about it. Then when you explain that to them you get got a pgi toe licker

topgun505, on 06 December 2014 - 09:48 PM, said:

*snip*

Equal skill? AYFKM? Game after game after game my teams seem to get rolled and half of my team (especially the assault pilots) and pulling in damages of 100-150 whereas the other side is pulling 500, 600, 700. I'm literally getting 2-4x the damage of the assaults in my Locust. Let that sink in a minute. A Locust is pumping out 4x more damage than those people. Once in a while with that happening ... sure ... no problem. Understandable. But game after game (like 20 in a row). I REALLY have a hard time with that above statement. It most definitely does NOT pit those of equal skill against each other.

....
i swear people...

Go back and read my post in this thread regarding the release valves and how it reaches further outside your Elo bracket to find teammates and opponents the longer it searches.
Then read the tips I listed on how to mitigate that and help prevent it.
Then please, please, PLEASE stop blaming the MM or implying that it doesn't do exactly what's been described.

Duke Nedo, on 06 December 2014 - 11:07 PM, said:

Ok mister, you really tick me off now. You blurt out "factual truths", when in fact you and me have access to exactly the same information. If you had been the coder of the MM you could make statements like that. Fact is, you don't know and I don't know anything more than what PGI has told us.. and that leaves out a lot of critical information. As long as we cannot use the information they have shared to fully reproduce the MM ourselves, they have not described how it works.

That does not make you enlightened or me uninformed. It just makes you arrogant and ignorant.

exactly
except I state the factual way that MM works as stated by the people who programmed it.
They HAVE described how it works
They even provided everyone with pictures on these very forums
Go do the research yourself. I'm tired of doing it for people too lazy to do it.

[Duke Nedo]

Sandpit, I have read what they posted about it. They leave out a lot of details. I don't complain that much about uneven games, that is something that is hard to avoid in 12v12. My main complaint is the streakyness of it. My point is that there is nothing, like you say as well, in the information they have provided that could in any way cause win/loss streaks. That makes me question what is going on in the MM.

I see streaks regularly, both wins and losses, and I tried to make the point that the frequency of observed streaks is higher than would be expected from random fluctuation, i.e. coin flips. Now, if you can accept that long streaks are statistical anomalies and it happens often, then it is natural to suspect that something is systematically wrong with the MM. The only other explanation I can come up with is that my mood is bipolar and flips between extremely destructive behavior destroying my teams chance to win, and extreme performance making our team a winning machine (actually even if remain afk in base, we still win). I switch polarity about every 20 games. I doubt that is it....

Edited by Duke Nedo, 06 December 2014 - 11:51 PM.

[Sandpit]

Duke Nedo, on 06 December 2014 - 11:50 PM, said:

Sandpit, I have read what they posted about it. They leave out a lot of details. I don't complain that much about uneven games, that is something that is hard to avoid in 12v12. My main complaint is the streakyness of it. My point is that there is nothing, like you say as well, in the information they have provided that could in any way cause win/loss streaks. That makes me question what is going on in the MM.

I see streaks regularly, both wins and losses, and I tried to make the point that the frequency of observed streaks is higher than would be expected from random fluctuation, i.e. coin flips. Now, if you can accept that long streaks are statistical anomalies and it happens often, then it is natural to suspect that something is systematically wrong with the MM. The only other explanation I can come up with is that my mood is bipolar and flips between extremely destructive behavior destroying my teams chance to win, and extreme performance making our team a winning machine (actually even if remain afk in base, we still win). I switch polarity about every 20 games. I doubt that is it....

and what I'm pointing out is that there's no code in the MM system that puts players into a "losing streak" as some in this thread implied.

I laid out exactly how the MM system works. They've provided pictures of the Elo spreads. They've explained it several times. It's not a complicated process.

MM simply places you into a game based on your Elo, weight class, and game mode selection.
That's it. Nothing more.
After a certain period of time searching for a game it hits a release valve which expands the Elo difference available. After another period of time it expands that range again so on and so forth.

That's all there is to it.

I gave tips on how to mitigate that if you find yourself getting placed into games you feel our outside of your personal range in this thread.

There's no other mechanic at work. It's been explained in writing by PGI and was even explained by Russ at the first townhall meeting. No amount of "well I don't believe it" changes the factual information provided by the people who programmed it. I'm sorry if player's "gut feeling" isn't substantiated by the factual evidence provided from the programmers.

If you find yourself in those types of "ruts", then go back and employ the tips I gave earlier in this thread.

PGI can't "fix" the MM system because it isn't "broken" to being with.

Now if you want to talk about ways that the MM system could be improved, that's a whole different ball of wax from "The MM sets up players to lose and employs some sort of coding that makes sure they lose if they win too much"

[Duke Nedo]

Again, in the best case only the programmer really knows that. Code can have bugs or unintentional side effects you know. PGI (certainly PGI some would say) are not flawless when it comes to coding...

I am not saying that the intentionally coded streaks into the system, but looking at the result coming out of the MM, is appears to not be working very well. The end measurable and quantifiable outcome is the facts. The information supplied by PGI is not facts, its information.

One example of something that makes me suspicious about the whole implementation: They do write that the Elo system predicts the outcome of the match and awards Elo for unexpected wins. Now, this imo is borderline an oxymoron in a 12v12 team game. The aim of the system is to create even matchups, ok? The prediction can only have any degree of accuracy IF the machup is uneven. You cannot by definition predict the outcome of a 50:50 chance of a win, it's a guess. Given the number of variables in a 12v12 matchup with individual players on lets be honest unbalanced maps, most people would agree that predicting the outcome would be extremely difficult unless the matchup is VERY uneven. Now, tell me that this doesn't bother you at all? From what I have read about the implementation of Elo and MM I'd say that if they want to claim half-even matchmaking they are better off just deleting the prediction part of the code and make all games count for Elo. Inaccurate prediction just increases the number of games needed to reach convergence.

Another point, if you look at the screens I posted from yesterday, tell me that the MM could not have done a better job dividing these 24 players into two more even teams....

Unless I have missed some info about the team-building, they have not described for example that they are using a final team scrambling before launching. From the sounds of it they just fill the team, release valves etc.

What they really should do to improve the balance is to select the 24 players, then divide them on teams using a monte carlo fitting of a scoring function that tries to describe player contribution based on Elo, BV and perhaps average reward2.0 over the last 100 games. What they then really really should do is to train that scoring function based on neural network feedback from the resulting match scores. That is my suggestion, then let everyone see exactly how it goes about it. Also, publish everyones Elo on the result screen including Elo changes as a result of the match.

[Sandpit]

Duke Nedo, on 07 December 2014 - 12:40 AM, said:

One example of something that makes me suspicious about the whole implementation: They do write that the Elo system predicts the outcome of the match and awards Elo for unexpected wins.

No sir, they do not.
Players write "The Elo system predicts the outcome of the match"
PGI write "Elo determines that players are in an evenly matched game with opponents in the same skill range"

[Vassago Rain]

Duke Nedo, on 07 December 2014 - 12:40 AM, said:

Again, in the best case only the programmer really knows that. Code can have bugs or unintentional side effects you know. PGI (certainly PGI some would say) are not flawless when it comes to coding...

I am not saying that the intentionally coded streaks into the system, but looking at the result coming out of the MM, is appears to not be working very well. The end measurable and quantifiable outcome is the facts. The information supplied by PGI is not facts, its information.

One example of something that makes me suspicious about the whole implementation: They do write that the Elo system predicts the outcome of the match and awards Elo for unexpected wins. Now, this imo is borderline an oxymoron in a 12v12 team game. The aim of the system is to create even matchups, ok? The prediction can only have any degree of accuracy IF the machup is uneven. You cannot by definition predict the outcome of a 50:50 chance of a win, it's a guess. Given the number of variables in a 12v12 matchup with individual players on lets be honest unbalanced maps, most people would agree that predicting the outcome would be extremely difficult unless the matchup is VERY uneven. Now, tell me that this doesn't bother you at all? From what I have read about the implementation of Elo and MM I'd say that if they want to claim half-even matchmaking they are better off just deleting the prediction part of the code and make all games count for Elo. Inaccurate prediction just increases the number of games needed to reach convergence.

Another point, if you look at the screens I posted from yesterday, tell me that the MM could not have done a better job dividing these 24 players into two more even teams....

Unless I have missed some info about the team-building, they have not described for example that they are using a final team scrambling before launching. From the sounds of it they just fill the team, release valves etc.

What they really should do to improve the balance is to select the 24 players, then divide them on teams using a monte carlo fitting of a scoring function that tries to describe player contribution based on Elo, BV and perhaps average reward2.0 over the last 100 games. What they then really really should do is to train that scoring function based on neural network feedback from the resulting match scores. That is my suggestion, then let everyone see exactly how it goes about it. Also, publish everyones Elo on the result screen including Elo changes as a result of the match.

The point of elo is to determine a winner and a loser.

People give PGI a lot of bad rep, but Karl Berg is a mad genius. His matchmaker is so good, it routinely produces 12-0 stomps - which is exactly what the elo is supposed to do.

If the matchmaker was perfect, every game would end in 12-0 or 0-12. There'd be no room for error. Each and every single match played would go precisely as the computer predicted.

So, you see, elo is really a no fun allowed score.

[Sandpit]

Vassago Rain, on 07 December 2014 - 12:56 AM, said:

The point of elo is to determine a winner and a loser.

see what i mean?

No Vass, it is not. The point of Elo is to give a player a number representation of their skill level in order to place them against other players in the same range of skill in order to have more balanced matches.
period.

[Duke Nedo]

Sandpit, on 07 December 2014 - 12:52 AM, said:

No sir, they do not.
Players write "The Elo system predicts the outcome of the match"
PGI write "Elo determines that players are in an evenly matched game with opponents in the same skill range"

Que????

http://mwomercs.com/...79-matchmaking/

This is an old post, but the prediction of the probable winner is central in Elo as far as I have understood this. That's how it changes Elo values after the match.

The bottom line is: Elo is really good for single or team ladder systems, It's about ranking players or team according to their ability to win. IF the matchmaker is doing a good job, both teams stand the same chance of winning, and thus the result cannot be predicted. There is no way to decide if the result should or should not reward a change in Elo.

If the aim of the MM is to create even matches and it can indeed predict the probable winner, then that match should not be launched. It should be rescrambled until no winner can be reliably predicted, then launched. Problem is, that breaks Elo.

So, it's actually not completely out of the blue to assume that PGI actually accepts a fair degree of imbalance because uneven Elo is needed to drive changes in Elo, and hence make Elo work. What I am trying to say is: if the outcome can indeed be predicted with any reliability, instead of launching that game and calculate Elo change from it, instead use that information to build a more even match.

[LordKnightFandragon]

Mystere, on 06 December 2014 - 08:59 AM, said:

Contrary to popular belief, the MM does not stack the deck. It's a perception problem, not reality.

If MMs dont stack the deck, then why in normal pug play are there literally no players with less then about 45% win and no players with over 60 some % WR? In WoT, the only ones with massive win rates are those who Clan wars, where they literally do stack their own decks and can be in a position where they literally cant lose. But in pubbing, you dont see even the absolute worst player with less then maybe 45% win rate and even really good players dont have more then maybe 55%. Only in very rare occasions do you see a player with less or more then that, and its just 1 or 2 players every like 20 games.

If the MM didnt stack the deck, then the good players, even in solo pubbing would be able to absolutely dominate the bads and would have higher then average win rates, while the bads would be 20-30% or worse Win rates. BUt its always right around 45-55%. Ive sat and watched my dad in WoT over 13K battles using the XVM mod, and the trends are a disturbingly obvious. He gets a really good streak, where his team gets the greens, the unicums and the deck gets stacked in his favor, he has to do little to nothing and wins just drop in his lap, then like the last few days, its been him, the only green, then rows and rows of terrabad reds. Record the last few days is like 6/30+.

MM Doesnt stack the deck...riiiight.

[Charronn]

Well all I can say is over the past 3 weeks or so I have been -35 games.I'm playing the same mechs,I'm playing the same way but the mm has done SOMETHING in order for me to be this amount of games down.
Conspiracy theory or not I have not become a bad player in that time and I have been expected to carry nearly every game with being in the shite team.I feel its doing it's best to get me down to 50% or less winrate.

Edited by Charronn, 07 December 2014 - 06:27 AM.

[Joseph Mallan]

Charronn, on 07 December 2014 - 06:27 AM, said:

Well all I can say is over the past 3 weeks or so I have been -35 games.I'm playing the same mechs,I'm playing the same way but the mm has done SOMETHING in order for me to be this amount of games down.
Conspiracy theory or not I have not become a bad player in that time and I have been expected to carry nearly every game with being in the shite team.I feel its doing it's best to get me down to 50% or less winrate.

But what are YOUR numbers looking like during those Losses? Are you doing well and the TEAM loses? Thats not the Matchmakers fault its attrition.

[Apnu]

LordKnightFandragon, on 06 December 2014 - 12:57 AM, said:

WoT uses your win rate, if you win like 10 in a row, you will be assured that your next 10 games will be absurdly bad. .

I already have this problem in MWO. If I have a good streak, MM punishes me with a bad streak the next day. If I have a bad streak, MM punishes me with more bad teams.

[Charronn]

Joseph Mallan, on 07 December 2014 - 07:12 AM, said:

But what are YOUR numbers looking like during those Losses? Are you doing well and the TEAM loses? Thats not the Matchmakers fault its attrition.

Generally I do near or most damage in the team.Always in the top half anyway put it that way.I have been marked for death by the mm lol.Time for a break maybe.

[Joseph Mallan]

Charronn, on 07 December 2014 - 07:55 AM, said:

Generally I do near or most damage in the team.Always in the top half anyway put it that way.I have been marked for death by the mm lol.Time for a break maybe.

Sounds like you are that player that is put into a Match cause there are not enough players available in your Elo at that moment. It happens. That is the Match maker doing what it can to get 12 on 12!

[Apnu]

Joseph Mallan, on 07 December 2014 - 07:12 AM, said:

But what are YOUR numbers looking like during those Losses? Are you doing well and the TEAM loses? Thats not the Matchmakers fault its attrition.

During the Victors challenge I tracked my stats accross 60 games. Took 56 matches to unlock the Victor. I had a KDR of 2.09, 550 average damage per match -- all in the same mech, the HBK-4J in the solo queue. I did that to test the MM vs my skill.

MM needs work. Or we need lobbies based off Elo rating.