344 replies to this topic

[Too Much Love]

slide, on 20 April 2017 - 10:29 PM, said:

What I think is happening most of the time (particularly during low population times) is the match maker is grabbing 12 players and making a team (because they have been waiting). It then grabs the next 12 players that hit launch, and because the valves have opened already, it launches the match. It would be similar to taking the above 24 players and putting in a teams split 1-12 and 13-24. Which would give an entirely different out come to the match up.

A lot of people think that it works that way, but the fact its - we don't know how it actually works.

All we have is the result of this work, that I tried to represent in my small study.

[SQW]

drunkblackstar, on 20 April 2017 - 11:38 PM, said:

A lot of people think that it works that way, but the fact its - we don't know how it actually works.

All we have is the result of this work, that I tried to represent in my small study.

But your data doesn't show how the MM allocate players to each team. The way how W/L, K/D and MS are closely related to each other, the 3 category should really be just be W/L.

Basically, your data just showed the better team ended up....winning more - not exactly ground breaking revelation. Sure, a team with 2 or more ADV will win 90% of the time but what's the probability of you end up on the 'winning' team? Where's OP in those 12 matches?

The outcry over MM stems from players who believes they should be on the winning team but got cheated because MM stuck them with 11 potatoes.

[Too Much Love]

SQW, on 21 April 2017 - 12:26 AM, said:

But your data doesn't show how the MM allocate players to each team.

Quite the opposite, it precisely shows how the players are allocated. Of cource, it could be the effect of some other principle of allocation, but they are allocated specifically the way I've shown. The result is the same.

SQW, on 21 April 2017 - 12:26 AM, said:

The way how W/L, K/D and MS are closely related to each other, the 3 category should really be just be W/L.

Why? It's your guess? You believe that W/L is more important etc.? It's no more then your opinion. How do you know how it should be?

SQW, on 21 April 2017 - 12:26 AM, said:

Sure, a team with 2 or more ADV will win 90% of the time but what's the probability of you end up on the 'winning' team? Where's OP in those 12 matches?

If you bothered to look at data I provided you would find this.

SQW, on 21 April 2017 - 12:26 AM, said:

The outcry over MM stems from players who believes they should be on the winning team but got cheated because MM stuck them with 11 potatoes.

No, it stems from those who want to understand how matchmaker works and why it makes people to lose or win in loopside streaks.

[slide]

drunkblackstar, on 20 April 2017 - 11:38 PM, said:

A lot of people think that it works that way, but the fact its - we don't know how it actually works.

All we have is the result of this work, that I tried to represent in my small study.

Actually I think your figures pretty much confirm what I said in the part you quoted.

One thing we know with absolute certainty, is for the most part MM doesn't work as intended, or as it should (if programmed wrong) simply because it does not have enough candidates to work with,

[Greyboots]

drunkblackstar, on 19 April 2017 - 09:06 AM, said:

I've reinforced my impression that the outcome of the match is determined by the matchmaker. In fact matchmaker doesn't assemble the equal teams. It makes teams to be unequal. The one team is determined to win, the other - to lose.

If you are going to play oline games involving PvP then get used to it.

This is why:

The general theory is that top end players don't mind losing in PvP 25% of the time. The matchmakers are therefore built to make these teams win to keep these players happy and their stats skyrocketing. At the smae time, those playing for fun instead of for the win don't really care all that much and are happy to lose 75% of the time as long as they get the occasional win. This is uaully achieved by putting people in the "will probably win" team to achieve the 25% win rate. .

It's not about fairness, it's about keeping people happy.

The erroneous assumption you've made is that it was ever supposed to create 'fair' teams in the first place.

So here's an example in action: Tier is an XP bar that goes up further than it goes down. It's balanced to make players progress no matter what. Play long enough and anyone can get to tier 1. If it was supposed to be fair it wouldn't do that, nor would it be biased by the performance of the team you're in but only go on personal performance to move up good players while keeping the others down in the lower tiers.

No, it's not because of the low population. The low population just makes it a lot more noticable because it gets hard for the matchmaker to make closer games so you get a lot of 12:0 roflstomps. Yes, it's all rigged.

Not a word of a lie, they teach this kind of stuff in many game design courses.

[Almond Brown]

drunkblackstar, on 20 April 2017 - 11:38 PM, said:

A lot of people think that it works that way, but the fact its - we don't know how it actually works.

All we have is the result of this work, that I tried to represent in my small study.

What is always missing in these "small studies" is the actual Pool of players available to the Match Maker at the time of construction. Time of day?, Day of the week?, Server played on? all would be MAJOR factors in the resultant picks from what you have seemed to "assumed" was optimal. We know just from these Forums that different Servers have different Pop values across the variety of TZ's on a daily basis etc.

As noted above, thanks for the attempt, but without the FULL set of parameters, especially the selection Pool available at the time of the Team selections, any data set, best of all from a very small set (12 units) is sketchy at best to try and gain a true understanding of how the MM works...

[Ghogiel]

MischiefSC, on 19 April 2017 - 07:54 PM, said:

You're only 8% of your teams value, approximately. So if you're a 5 W/L in pug queue (winning about 83% of your matches, or 5 out of every 6) but your team of 11 is all at a 1.0 (winning 1 out of 2 matches, or 50%) then your team average W/L is 52.7%.n it's still going to end up unbalanced often enough that it wouldn't be worth the effort.

I think this is what the OP is going to be looking at more often than not>

12 players with 1:1 W/L vs = avg 1

vs

11 players with 1:1 with the 0.5%er ringer with 5 W/L = avg 1.33

30% difference in avg W/L between the teams.

[Escef]

drunkblackstar, on 20 April 2017 - 11:19 PM, said:

For all those guys who SUDDENLY became an experts in statistics:

1) I knew that there would be comments like that (though I didn't expect that there are so much statistics professors playing MWO), so I made the special remark concerning this problem in my original post. Apparantly, it didn't help. Nobody reads the post till the end, I know.

Why read past the point where you said your data is insufficient? Your "study" is worthless.

[Too Much Love]

Almond Brown, on 21 April 2017 - 06:19 AM, said:

What is always missing in these "small studies" is the actual Pool of players available to the Match Maker at the time of construction. Time of day?, Day of the week?, Server played on? all would be MAJOR factors in the resultant picks from what you have seemed to "assumed" was optimal. We know just from these Forums that different Servers have different Pop values across the variety of TZ's on a daily basis etc.

As noted above, thanks for the attempt, but without the FULL set of parameters, especially the selection Pool available at the time of the Team selections, any data set, best of all from a very small set (12 units) is sketchy at best to try and gain a true understanding of how the MM works...

Imagine cooking machine.

This cooking machine makes 12 biscuits. Sometimes 8 buscuits are burnt, sometimes 9, sometimes 10.

After 12 times you used it, it would be safe to say that this cooking machine makes burnt biscuits.

But instead you are confused. You feel that its important to answer several questions before stating that it is a burnt buscuits machine.

"Time of the day? Current voltage in the network? Dough used?"

Edited by drunkblackstar, 21 April 2017 - 06:44 AM.

[Ghogiel]

drunkblackstar, on 20 April 2017 - 01:15 PM, said:

But the very problem is that the teams are NOT equal. They are created unequal.

How would you know or verify that they aren't equal when the single variable that is used to create the teams you haven't even looked at/can't look at? You haven't listed the PSR tier of any of the players so for all you know the teams might be exactly equal in all of your data collection.

I suspect they aren't exactly matching, but you cannot in anyway conclude that the teams are unequal until you look at the only relevent data that is used by the MM to determine if the teams are equal or not.

[CDLord HHGD]

Only read the title...

Are you new here? The MM has been FUBAR since day one.

[SQW]

drunkblackstar, on 21 April 2017 - 12:44 AM, said:

Quite the opposite, it precisely shows how the players are allocated. Of cource, it could be the effect of some other principle of allocation, but they are allocated specifically the way I've shown. The result is the same.
Why? It's your guess? You believe that W/L is more important etc.? It's no more then your opinion. How do you know how it should be?

If you bothered to look at data I provided you would find this. No, it stems from those who want to understand how matchmaker works and why it makes people to lose or win in loopside streaks.

So you played 13 random games and won 6 and lost 7. The team with the best overall K/D, W/L stat (so basically the better team) is more likely to win. I still don't see the problem with MM.

I think I have over a thousand completely random solo QP games under my belt and I've never had winning or losing steak more than 6/7 matches long - usually no more than 4. I don't get what your perfect MM is suppose to achieve. No more stomps? Nail biting finish every time? If I record all my games, the match score (in team kills-deaths) would be a flat bell curve with a few stomps/wipes on the extreme end and majority falling in between.

[Too Much Love]

Ghogiel, on 21 April 2017 - 06:55 AM, said:

How would you know or verify that they aren't equal when the single variable that is used to create the teams you haven't even looked at/can't look at? You haven't listed the PSR tier of any of the players so for all you know the teams might be exactly equal in all of your data collection.

I suspect they aren't exactly matching, but you cannot in anyway conclude that the teams are unequal until you look at the only relevent data that is used by the MM to determine if the teams are equal or not.

I was tempted to look at Tiers, but there are several problems.

The first one is it requires much effort. I need to find the player at the forum, etc. The second - not everybody show their tier. So it's quite difficult to accomplish.

Besides, they recently confirmed that due to low population they loosen the tier restriction. Before Tier 1 was matched only against T1-3. Now it can be matched against almost anybody. So the necessaty to find tiers is not that high.

Moreover I suppose that matchmaker takes into account much more then tiers numbers.

[Ghogiel]

drunkblackstar, on 21 April 2017 - 07:23 AM, said:

So the necessaty to find tiers is not that high.

Moreover I suppose that matchmaker takes into account much more then tiers numbers.

If you wanted to find out if the teams are equal according to the MM, then all the players PSR would be entirely essential and the only info you would need. The stats have almost nothing to do with PSR and MMing. Just because there is a tier spread in the MMing doesn't mean the MM isn't doing a good job of matching the tier spread between teams, I personally doubt it but it could be working as intended and within threshold (which could also be loose af but you have no idea whether the teams are equal or not and never will know unless you know all the players tiers in a given match)


You will probably suppose wrong to think that the MM is taking into account stats other than PSR, to think otherwise would be contray to the dozen of explaination from karl berg threads, town halls and command chairs on the subject. That sort of thing is nearly tinfoil assertions

[Mister Glitchdragon]

This settles the question regarding whether its harder to maintain high stats in Group or Solo queue.

Based on this finding, it proves players who only drop solo will have a harder time maintaining high stats, since the MM will actively hobble their efforts by not only pitting them against good players, but also by dropping them with outmatched teammates.

Players who drop group, on the other hand, will have an advantage because they can control that variable, with 12-mans having the greatest MM advantage, regardless of ability, since the matchmaker will never be able to hobble their team.

Good to know.

[MrJeffers]

Ghogiel, on 21 April 2017 - 07:46 AM, said:

If you wanted to find out if the teams are equal according to the MM, then all the players PSR would be entirely essential and the only info you would need. The stats have almost nothing to do with PSR and MMing. Just because there is a tier spread in the MMing doesn't mean the MM isn't doing a good job of matching the tier spread between teams, I personally doubt it but it could be working as intended and within threshold (which could also be loose af but you have no idea whether the teams are equal or not and never will know unless you know all the players tiers in a given match)


You will probably suppose wrong to think that the MM is taking into account stats other than PSR, to think otherwise would be contray to the dozen of explaination from karl berg threads, town halls and command chairs on the subject. That sort of thing is nearly tinfoil assertions

Exactly.

The MM works on three criteria that we know of - time, PSR rating, and Mech class. That's it. Time is only used for two things - to seed the match with the player/group who has been in queue the longest, and then as a timer to open release valves on Mech Class restrictions and then PSR rating.

It fills the teams in an alternating fashion, adding player or players in the case of group queue that match the selection criteria to the team with the fewest players until both teams are full.

It doesn't know what the ending set of 12 players on either team will be, and it doesn't change the makeup of the teams as they are created.

Edited by MrJeffers, 21 April 2017 - 08:08 AM.

[vandalhooch]

drunkblackstar, on 19 April 2017 - 10:09 AM, said:

Yes, this model works for Solo queue primary. The group queue waits for it researcher:) IMHO it works almost the same way.

The devations to the stats and to the outcome are possible (grinding mech, pumped stats etc.), but they don't affect the general mechanics of matchmaker.

But the possibility of group queue stats mixed in with solo queue stats invalidates your entire "study" and "analysis" of twelve non-random matches.

Basically, your "study" is nothing of the sort. Mark Twain rides again.

[vandalhooch]

MischiefSC, on 19 April 2017 - 11:59 AM, said:

40 is a solid statistical sample but 12 isn't terrible. Good enough for a "guidance" look.

I'm just wondering if the virtual lobby can shuffle. As in once it pulls someone to put in a lobby does it shuffle teams or rebuild lobbies.

12 random matches might be interesting but there was absolutely nothing random about the matches "analysed" here. They were all consecutive, meaning a short window of time for a game played globally. More importantly, the analyzer himself is in ever one of the matches. Absolutely nothing random about that, not to mention the observer bias created by his knowing that he was going to screen shot and "analyze" the data afterwards.

[MischiefSC]

vandalhooch, on 21 April 2017 - 02:45 PM, said:

12 random matches might be interesting but there was absolutely nothing random about the matches "analysed" here. They were all consecutive, meaning a short window of time for a game played globally. More importantly, the analyzer himself is in ever one of the matches. Absolutely nothing random about that, not to mention the observer bias created by his knowing that he was going to screen shot and "analyze" the data afterwards.

Sure there's bias - however that would only impact OPs behavior. The value of the data is in the stats of the other 23 plays and what their relative w/l was.

The data wouldn't be worth a fart in a mitten to evaluate OPs skill or performance. However to evaluate what the relative w/l of all the players on both teams was?

It's good enough for general guidance. I wouldn't use it to build a business case but I'd use it to determine if there was a basis for doing a full study to make a case from. It's enough for me today "that's statistically significant and there's some viable logic to build an argument about".

[vandalhooch]

drunkblackstar, on 20 April 2017 - 01:23 PM, said:

May be. But

1) Who really knows how it works? How can we discuss it, if we simply don't know. It's all rumors + some little pieces of information PGI made public.

And your analysis is based on . . . a non-random sample size of twelve matches?

Quote

But know we can analyze the comparision of teams in detail, that is the result of MM work.

2) Your words don't explain the things like streak-defeats or streak-victories. If the teams were randomly picked up by tiers only, they would be more or less even chance of defeat/victories. The situation of 10 defeats-stomps in a row wouldn't be possible.

Aaaaannnd you're done. Anyone who's statistical knowledge is so limited that they think the above statement is true should not be trying to "analyze" anything.

Streaks happen. Even in truly random systems, streaks happen. Pointing to occasional streaks as evidence that something is not random (i.e. a 'rigged' matchmaker) is ridiculous. You need access to the results from thousands and thousands of matches to show that long streaks are more or less common than predicted by chance. Without that data, any individual streaks you observe, including Taragato's cherry picked nonsense, is simply an observation.

Quote

Why are you wasting your time writing this post then? You better enjoy gaming.