140 replies to this topic

[Duke Nedo]

Note: this is based on dropping solo 100%

This is a plead to PGI to remove some of the fog surrounding the MM. There have been lots of threads lately about poor matchups and abnormal MM behavior (streaks). Since noone really knows exactly how this works it all comes down to guesses and manure-flinging... so why don't you help us out? It could not be worse in the eyes of the public and you might actually get some useful feedback from the community (joke intended ). I assume that the people that want to abuse the system to sync have already discovered any playable parameters so you might as well just open up that black box to everyone.

Personally, I have experienced abnormal behavior that I blame the MM for, that is loss streaks that feel very unnatural. When this starts you know that you are in for it for a while, all you can do is to try to mitigate the damage (carry harder, alright!). The problem is the strong feeling that this is systematic and not caused by random chance.

For example: I mastered by Victors before Clan invasion. Post agility nerf, pre jj nerf. I experienced what I would call normal matchmaking, some wins, some losses, no major loss streaks longer than say 3-4 games that you would expect to happen frequently from random chance alone.

VICTOR VTR-K 60 34 26 1.31 49 38 1.29 16,371 58,338 06:06:45
VICTOR VTR-9B 57 28 29 0.97 50 37 1.35 16,140 57,975 06:12:31
VICTOR VTR-9S 69 41 28 1.46 74 36 2.06 22,329 78,714 07:30:39

A win/loss ratio of ~1.24 over almost 200 drops, which I think is pretty decent. I tried to give some pointers to team tactics and carried as hard as I could, without hiding behind team mates etc. Nothing fancy, probably average elo, average damage/match, average k/d ratio for 100% solo dropping. Now comes clans, 3-3-3-3 and MM tweaks. Here are some examples from chassi that I tried to skill during loss-streaks:

TIMBER WOLF TBR-S 42 13 29 0.45 41 29 1.41 16,711 50,973 04:23:10
DIRE WOLF DWF-PRIME 36 13 23 0.57 37 24 1.54 15,261 38,339 03:35:02

Small sample size, yes, but the trend is there. Decent k/d ratios still of ~1.5, but a w/l ratio down to ~0.5. These two chassi both included ~10-match loss-streaks with matchups beyond anything near possible to carry, no matter what. 10 times in a row! That's where I think something is fishy with the MM. 10 loss streaks can happen, but the probability that they will happen to the same player several times over as little as 700 games (after stat reset) is just extremely low. And severa others have reported that same on the forums, and been called crybabies for it. So, flame ahead.... but I assure you, if you really look forward to a gaming evening with some beers its really, really frustrating to get impossible odds thrown at you 10 times in a row. I rather watch a movie.

Over all I am still slightly on plus, so you may say working as intended:
Wins / Losses 382 / 359

but I disagree. My k/d ratio is increasing, meaning I am improving and contributing more for every game I play, but m w/l ratio is plummeting. I used to be +70 wins over losses and during the two last weeks or so that as been reduced to +23 now. So, losing a ~50 games advantage over the course of say 200 drops really hurts so that you can feel it!

Now, if ELO and MM was working as intended, I would expect my w/l-advantage to stabilize after I reach "my elo", not to be dragged down. It would imply that my elo suddenly got heavily inflated for some weird reason. I don't believe that. I do however have a strong feeling that something systematic is at play, and I would pay real money to be able to avoid unnatural loss streaks. Another personal and circumstantial observation is that I dont have a single win-streak for more that 6-7 games, not a single one. My w/l ratio is built up over time, then I suddenly get several looong loss streaks. That adds to the impression that this is unnatural.

I plead to PGI to clarify a few things like:

* Sync drops - these could cause loss streaks in the solo queue, the mechanism is there but how common is it? I believe this is a boogieman, it can't possibly be that heavily abused... but it could be that it gets the blame for an actual real systematic behavior in the MM. Not sure if there is a simple way to collect statistics, but would feel good to be able to kill thay boogeyman.

* 3-3-3-3 - This could possibly contribute to loss streaks depending on how matches are built. Circumstantial feeling tells me that I am less likely to get streaks of bad teams if I select a class that is played less. No way of separating this from an ELO effect though since my elo will be different for different classes.

* ELO - Been discussed a lot. But can it cause loss streaks? Probably. For example, I have got more loss streaks when there is a new package released or a weekend challenge. That means more people from the group queue injects into the solo queue. They are likely to have abnormal elo from playing in groups (good or bad) relative to solo player elo. That is even probable to screw someone over....

* Spawn points - Could pgi please clarify if spawn selection is completely random? How are the spawns filled relative to how the match is built relative to elo etc? Once again a feeling, but most maps are imbalanced and getting the bad side greatly reduce the chance to win.... and in the loss streaks you usually repeatedly get the bad side. Big chance that this is biased perception, but mentioning it anyways. At least I believe this GREATLY contributes to making otherwise even matchups end up in steamrolls. Tourmaline, alpine and crimson are the worst imo.

All in all though I think elo somehow works, though you have to play many many many games in order to reach a proper elo level, and then it is probably very sensitive to inflation from playing mixed in group or solo queues. Anyways, I very rarely see complete beginners or trial mechs anymore and many matches have been Good lately, both wins and losses. The thing I struggle with is only the streaks. That said, scrapping elo is probably a good idea. 12 v 12 is too big for elo, it almost turns random at some point until you have played hundereds and hundreds of games... for each class.... I would much prefer experience brackets of some kind, it would give bigger player pools inside each bracket and search times would go down.

Please? Open the black box! What's in the booooox *insert Mel Gibson Brad Pitt here*

Edited by Duke Nedo, 02 September 2014 - 01:09 AM.

[Kyrie]

Duke Nedo, on 02 September 2014 - 12:58 AM, said:

What's in the booooox *insert Mel Gibson here*

While completely off-tangent to your post, I always associate the "whats in the booox" line with the movie "Seven", and Brad Pitt.

[Duke Nedo]

Lol, Brad Pitt it was... getting old!

[El Bandito]

I too want more clarity on PGI's end. However, my WLR is still having similar ratio from pre-Clan era. I have no problem with the current MM. Especially since groups are now gone from the solo-q.

[MrMadguy]

I experience long lose streaks very often. And I found two regularities about when it happens the most: 1) When activity on servers is very low - chance to get lose streak is much lower, when we have some tournament or challenge. So, lose streaks could be connected with too narrow range of players to choose from. Also I've noticed, that MM tends to choose big "waste of time" maps like Therma and Alpine at such a moments. 2) When I'm leveling new mech. I don't know, why it happens. I don't think, that it's due to new mech underperforming, cuz in most cases I change loadout and make all necessary updates before starting to level my mechs up. The fact is - when I'm leveling new mech it may take half a day to get 2x bonus on it due to non stop losses. And other interesting thing: I start losing not only while playing this mech, but also on all other already leveled variants of that mech. It really feels like devs want to force me to pay for premium time or exp transfer.

[Joseph Mallan]

24 go in 12 or less come out!

[Triordinant]

Joseph Mallan, on 02 September 2014 - 03:11 AM, said:

24 go in 12 or less come out!

Two teams enter... one team leaves!

[John1352]

The key is, don't carry too hard. The limited information we have on the matchmaker/ELO system is basically this: (I am not going digging for references, go look yourself)

http://en.wikipedia....o_rating_system

• Each player has a light ELO, a medium ELO, a heavy ELO and an assault ELO.
  
• MM makes a game.
  
• MM predicts a winner based on which team's total ELO is higher.
  
• If you are the predicted winner, and you win, your ELO doesn't change. Same if you're the predicted loser and you lose.
  
• If you are predicted to lose, and you win, your ELO increases. If you are predicted to win, and you lose, your ELO decreases.

Here's where it gets contentious: Players are supposed to be put into "ELO buckets" This means that good players who win a lot would only usually face other good players. The general forum consensus is that this doesn't happen. Instead, the matchmaker only tries to give the teams the same ELO, for example team A gets one good player, and one bad player, team B gets two medium players. This leads to the idea that the better you play, the worse the team you get put on is.

[Too Much Love]

Duke, I have the same problems.

I don't blame synch-drops, there can nit be that much of them to spoil the game.

it's pure matchmaking system fail.

You're right you're becoming better, your kd ratio grows.

But this system not only doesn't encourage self improvement, it is punishing it! Paradoxically, but the better you are, the worse your team. It compensates your skill by the low level of your teammates to the point that majority of your team wi die in the first few seconds of the match. At least, this is my guess, because it's hard to say when all the data is hidden.

[Duran Vancor]

I can share my experience from today. Had some time on my hand and played lots of matches. The first 13 were all loses, my biggest losing streak so far and a new record.

My personal performance was irrelevant. 800 damage, 5 kills, 3 UAVs destroyed? Lose. Suicide run into the enemy and die with 5 damage 1 minute in? Lose. The last matches I played were better, with wins and loses, but I have to admit, they were never fun.

What I noticed is, that the matchmaker does not differentiate between for example a Dragon and a Timberwolf. So 3 Timberwolfs on one team vs 2 Dragons and an Orion or similar happened quite often.

On a side note... the trial Nova is my second best performing mech but has the lowest win/lose ratio of all with 0.33, which is hilarious!

[Faith McCarron]

I think the best example of this was Jager's stream last week during the final day of the 50 win challenge. He was solo pugging on the stream to get the last 25 wins, and it was painful to say the least. He was carrying as hard as ever, dropping in good builds, all of that, and yet he was regularly running loss streaks of 5-10 games. In a normal night you figure it would take Jager what, 30-35 drops at most to get 25 wins? I dont know how many it ended up taking, but it was well over 50. It was just giving him garbage teams, no matter how many kills, how much damage he did, how many openings he made for the team, how many enemies he tied up, his teams were just so terribly bad that a win was simply impossible.

[Livewyr]

Faith McCarron, on 02 September 2014 - 04:34 AM, said:

I think the best example of this was Jager's stream last week during the final day of the 50 win challenge. He was solo pugging on the stream to get the last 25 wins, and it was painful to say the least. He was carrying as hard as ever, dropping in good builds, all of that, and yet he was regularly running loss streaks of 5-10 games. In a normal night you figure it would take Jager what, 30-35 drops at most to get 25 wins? I dont know how many it ended up taking, but it was well over 50. It was just giving him garbage teams, no matter how many kills, how much damage he did, how many openings he made for the team, how many enemies he tied up, his teams were just so terribly bad that a win was simply impossible.

I knew that experience all too well.

[Wolfways]

Here's how the MM works most of the time from my POV (100% solo drops)

It picks out 11 players with the least experience, worst builds, using joysticks, preferably using 3pv....then puts them on my team.

[Duke Nedo]

Thanks for adding your experiences here!

What I am very interested in learning is not how elo works, but how the MM actually goes about sequentially to seed a team - what elo does it start from, which variance does it look for, which classes in what order, how it fllls the slots - which criteria are used to select from the available pool, then when it gets stuck, which restraints are being slacked in which order and then which criteria are used etc. How is it decided who goes on which side? Are lances built together as unit or do the end up as they land on each side?

How many matches are filled simulateously? Are they, or are they built in sequence?

I think that could kill a few boogiemen and pin down any anomalies...

Edited by Duke Nedo, 02 September 2014 - 05:03 AM.

[LowSubmarino]

Faith McCarron, on 02 September 2014 - 04:34 AM, said:

I think the best example of this was Jager's stream last week during the final day of the 50 win challenge. He was solo pugging on the stream to get the last 25 wins, and it was painful to say the least. He was carrying as hard as ever, dropping in good builds, all of that, and yet he was regularly running loss streaks of 5-10 games. In a normal night you figure it would take Jager what, 30-35 drops at most to get 25 wins? I dont know how many it ended up taking, but it was well over 50. It was just giving him garbage teams, no matter how many kills, how much damage he did, how many openings he made for the team, how many enemies he tied up, his teams were just so terribly bad that a win was simply impossible.

If you have a significantly worse team then your individual performcance simply cannot compensate for other players doing really bad things like spreading all over the map. They will all die. Or a wild cap-chase on the smallest of maps instead of simply staying in the group and killing the other team.

I agree. That is pretty frustrating because in 9 out of 10 times when you are 3 mechs down 2 minutes into the game you already know that the game is already lost no matter what you do. I can literally predict how games end after 1 or 2 minutes. All you need is to see who loses the first few mechs and glance at the map. The way your team is positioned provides enough information to predict who will win.

Since I frequently know how games will end 1 - 2 mins into the game, no matter how good I play - I can fight like a berserker or shut down - I sometimes just lose interest in that game and switch to anothe mech. There is no fun if the game is over too early.

What I dont understand is how ppl can say that the mm works. It doesnt. At all. How can it workt when I - played for 1,5 years - with reasonable skills get dropped with absolute bloody beginners that seriously suicide in the most ridiculous ways? To then play against a team with players that all play like me or even better?

And the next game I am in the team with skills and we simply annihilate the other team 12:1. 12:4, 12:5 or 12:6 would be somewhat ballanced. The other team at least managed to extract a price and put up a somewhat effective fight.

But those complete annihilations with teams that clearly do not share the same skill level....that just isnt fun. And that is why I too get a lot of streaks with 10 losses or 10 wins in a row. The players are mixed up badly. So badly, that there is no chance whatsover for that 'underskilled' team to win. Even if the skilled team all played in locusts. Theres a good chance they would still win. The skill difference is simply too huge. There are ppl that apparently either never played any pc/video games in their life, are 6 years old or simply fail to understand some of the more basic strategies and tactics of such a game.

And im not raging against those ppl. Not at all. To them it must be utterly frustrating to play against ppl that just destroy them, find them no matter where they are on the map ("oh....theres a map?") and they barely got to shoot their guns because they overheat every thrid volley haha.

No, I could provide dozens of more examples. I even got 15 win or loss streaks sometimes. The matchamker doesnt work and I dont care if 99,9 % ppl in this forum reject that impression. I have played a lot again the last month and I saw exactly what kind of ppl played against each other in those games.

Whoever says the matchmaker works - or even partially works - plays another game than me. The mm is badly designed, does not even remotley fulfill its purpose of pitching similarly skilled ppl against each other and also throws absolute beginners, newbs and noobs against very skilled players.

There are some more balanced games too. But they are a rare treat. I would say every fifth game is somewhat balanced and every 10th game is truely balanced. That is not enough as I often stop playing for the day not because I dont wanne play anymore but because I got bored after predictable and totally unbalanced games.

Maybe it's too complicated or even impossible to design a somewhat working mm. But one thing they really should be able to correct is to mix absolute beginners against experienced players.

[Zolaz]

I think even PGI is afraid to touch their own MM with a 10 foot pole. Remember how many patches it took to get the current MM we have?

[Willard Phule]

Zolaz, on 02 September 2014 - 05:52 AM, said:

I think even PGI is afraid to touch their own MM with a 10 foot pole. Remember how many patches it took to get the current MM we have?

Keep in mind that the current MM we have is a re-written version that came out around the same time the Clans came out. There's something about it on Twitter that got quoted here somewhere...but, the gist of it is that the better you are, the more new players you have to carry. It's gotten seriously out of hand lately.

[Greenjulius]

I'd also like to put my 2 cents in. I almost always drop solo. Occasionally I'll drop in with one other friend.
During the 50 win solo challenge, I attempted to use heavies and assaults and found that no matter how hard I played, I lost 4/5 games. I got frustrated after being unable to win a match in my Jester. (worthless mech) Probably 12 drops in it, decent games from me, but all loses. I threw my hands up and said, screw it. This game hates me and this mech and won't give me a decent team.

I tried my Kitfox and got a win. Then another win. And another win. I finally had a win streak, and it's because I stopped using mechs over 50 tons. Since then, I've decided the only way the matchmaker will give me a decent game is when I drop in with lights. I have to carry REALLY HARD but at least I have a chance. I just need to kill at least 3 mechs per game, and my chances of winning go up to 50%+.

To this day my Jester is the single most worthless mech I have. I have something like a 35% win rate and dropping. I really wish I could sell it for a fraction of the MC I lost on it. Even with the 30% Cbill boost on the Jester, I make more running a K2 Catapult.

So, what I'm saying is this: Fix the matchmaker. I won't spend a single dime more on this game until I have decent matches that don't force me to take stimulants to have a chance to win.

Edited by Greenjulius, 02 September 2014 - 07:22 AM.

[KingCobra]

I would suggest to PGI just turn of the MM AI and let players just be placed in matches with no ELO except new players for there first 50 games just match by tonnages only let the player play on skill alone. If you want to know how the MM really work just down load the cryeng3 SDK and play around with it.
http://www.moddb.com...gine-3-free-sdk

Here is a link to Cryeng3 tools and AI MM system you can look over.
http://cryengine.com/features/tools

AI SYSTEM


Realistically rendered and animated characters require state of the art AI systems to intelligently respond to the game environment and maintain the illusion of realism.

CRYENGINE features powerful, scalable and flexible AI technology to handle character behaviors with modular sensory systems (such as sight and hearing) and fully support the complex requirements of the character locomotion system. AI in CRYENGINE requires no additional middleware and is fully integrated for your game within CRYENGINE Sandbox, with WYSIWYP functionality on all platforms.

[Greenjulius]

I don't think the CryEngine has built in matchmaker options. The built in AI is for character AI, not match or server based logic. The matchmaker in use by MWO is almost surely a construct of their own. A monster without compare.