140 replies to this topic

[White Bear 84]

BLOOD WOLF, on 03 September 2014 - 05:36 PM, said:

My win/loss has pretty much stayed the same if not growing. As I am now winning more matches then I am losing. My Kills at death ration is 0.95

I can't really complain, I am at 2.15 which is generally pretty sweet for a pug/4 man cross...

Generally speaking though I find the decrease is relative to the number of losses. For example, mechs that have a W/L less that 1 are usually have a KDR of less than 2. Those with a W/L around 1-1.5 vary the most, usually sitting at around 2. This is the most common variance., while those with a W/L of > 2 generally always have KDR >2 , if not >3. Exceptions to this rule include role warfare mech such as tagging/uav/ecm lights.

Then of course you have to take assists into consideration. It is not always representative to have a low KDR but say consistently get high number of assists and decent damage... ...irrespective, you can easily track a drop in KDR back to a drop in W/L which usually matches up with a bad streak - characterised by consistent 12-3/2/1/0 matches.

Edited by White Bear 84, 03 September 2014 - 06:17 PM.

[BLOOD WOLF]

White Bear 84, on 03 September 2014 - 05:44 PM, said:

I can't really complain, I am at 2.15 which is generally pretty sweet for a pug/4 man cross...

Generally speaking though I find the decrease is relative to the number of losses. For example, mechs that have a W/L less that 1 are usually in the KDR of less than 2. Those with a W/L around 1-1.5 vary the most, usually sitting at around 2. This is the most common variance. Those with a W/L of > 2 generally always have KDR >2 , if not >3. Exceptions to this rule include role warfare mech such as tagging/uav/ecm lights.

Then of course you have to take into consideration, assists. It is not always representative to have a low KDR but say consistently get high number of assists and decent damage... ...irrespective, you can easily track a drop in KDR back to a drop in W/L which usually matches up with a bad streak - characterised by consistent 12-3/2/1/0 matches.

My Timberwolf prime has a below average K/D but an above average W/L.

1.22, which means I win 20% more than I lose in that mech. of course I can do a good amount of damage in that mech.

Edited by BLOOD WOLF, 03 September 2014 - 05:55 PM.

[White Bear 84]

BLOOD WOLF, on 03 September 2014 - 05:54 PM, said:

My Timberwolf prime has a below average K/D but an above average W/L.

1.22, which means I win 20% more than I lose in that mech. of course I can do a good amount of damage in that mech.

That might mean you are either lucky or the extra damage and work you are doing in that mech is contributing to your team winning.

That said, you probably have your fair share of players that could carry harder in your matches which would lead to that figure being even higher. It is a team game after all.

[BLOOD WOLF]

White Bear 84, on 03 September 2014 - 06:19 PM, said:

That might mean you are either lucky or the extra damage and work you are doing in that mech is contributing to your team winning.

That said, you probably have your fair share of players that could carry harder in your matches which would lead to that figure being even higher. It is a team game after all.

I agree with that. Always seems to be player on my team that do near 1000 almost every match, its insane. makes 500 look shameful.

Edited by BLOOD WOLF, 03 September 2014 - 06:22 PM.

[White Bear 84]

BLOOD WOLF, on 03 September 2014 - 06:22 PM, said:

I agree with that. Always seems to be player on my team that do near 1000 almost every match, its insane. makes 500 look shameful.

Not at all. 500 is a good score to get. If you get 1000 you could easily be destroying components left right and centre, but your 500 might equate to more accurate kills/assists...

Now, if you start drifting below 200....

(I wont even go into sub 100 territory - especially with heavies and assaults...)

[BLOOD WOLF]

White Bear 84, on 03 September 2014 - 06:38 PM, said:

Not at all. 500 is a good score to get. If you get 1000 you could easily be destroying components left right and centre, but your 500 might equate to more accurate kills/assists...

Now, if you start drifting below 200....

(I wont even go into sub 100 territory - especially with heavies and assaults...)

Like watching a direwolf pilot stare right at a mech and not hit the fire button.

[White Bear 84]

BLOOD WOLF, on 03 September 2014 - 06:43 PM, said:

Like watching a direwolf pilot stare right at a mech and not hit the fire button.

Or an Atlas just standing at 1000m with a full arsenal of LRM's just firing two LL.

Edited by White Bear 84, 03 September 2014 - 06:44 PM.

[Duke Nedo]

What I really am curious about with the MM is if there actually are systematic contributions that causes streaks of wins or losses. I am very aware of how random works and that some people will get long streaks... but I have a strong feeling that streaks are too common, from my experiences and from reading other people's stories. Streaks are not directly linked to uneven games, though uneven games are necessary to produce artificial streaks (if all games were even you would have 50:50 to win all games and any streak would be random). Now, question is not if there are uneven games, everyone knows there are and I could easily accept that if my chance of getting the bad team or good team was 50:50. It would be less fun and rewarding, but if it's 50:50 I would never feel abused.

If I do some low grade maths, the chance of a 10-matches loss streak is 0.5^10 = 0.097% chance. It happens. Over the course of my 3500 drops, lets say I have the chance of a streak ~700 times (assumed ~5 games in average are consumed by "failed streak attempts", I am sure someone better at maths will work out how to do that) that chance of seeing one such streak is 1 - 0.9990234^700 = 49.5% chance. So, I stand a ~ fifty fifty chance of having a 10-match loss streak if I play 3500 games and have 50% chance of winning every game. Given the number of players here, quite a few are bound to have several of them. The chance of a win streak is identical.

What I am curious about is if others have experienced this getting much much worse lately, since I cannot draw any conclusions based on my experiences alone, and judging from the fourms I'd say that many players are experiencing this right now in the solo queue. I have now had 2 such streaks in the last 400 games, the chance of that with the same reasoning is ~3-4%, so it can happen. Actually, one was a 13-game loss streak so less than that, but whatever. In addition I have also had many many loss streaks of 5-9 matches and only one single win-streak of 7 matches during these 400 games. My k/d ratio has increased because I have played rather ok by my standards, while my w/l ratio has dropped a lot the last two weeks. Really a lot, went from +70 more wins than losses town to +20, over as little as ~200 games. That's about 75 wins and 125 losses, trust me you can feel that. Add to that that way back when we did not have elo, matchups were horrible and stomps very common, BUT I personally never experienced any loss streak longer than 4-5 games or so. Nothing that made any impression.

Now, my personal contribution in a team game like this, 12 v 12, amounts to ~8% of the team effort. If I feel cocky I can claim that to be 20% or 30% or whatever. It doesn't matter, at the end of the day the team effort is dominated by the collected efforts of my teammates, which in turn means that I am basically at the mercy of the MM. Therefore I conclude that my own efforts are likely not to blame. The 50-wins challenge proved that point where people in lights would quickly suicide and then go to the next game and repeat. Their win/loss ration was perfectly fine...

These arguments goes for the ELO as well and that is one of the major problems imo. Personal ELO based on team performance where your individual contribution is <10%. Same would go here actually, the MM will dominate my ELO change in every game. The idea then is that the dominating contribution from the MM is supposed to cancel out (your teammates) and only your contribution will remain after x nr of games. That is bound to leave artifacts, wiggles on the baseline. You would have to makes hundreds and hundreds of drops for every class to get perfect cancellation of your teammates contribution to your elo. Now, this is in a perfect world where everyone drops solo, preferrably in the same loadout, same mech within the class.... I fail to see how elo is a good idea.

Atleast for the solo queue I would really like the MM not to use ELO. Use experience brackets of some kind. Say:
Bracket 1. Greenhorns < 3 mastered mechs
Bracket 2. 3-12 mastered mechs
Bracket 3. 13+ mastered mechs

Well... I am just curious. It could be that I am just really, really unlucky, but since others are seeing similar things in the solo queue I am not so sure.... extra curious if I will bounce up again after I have dipped 1.0 w/l ratio, it will happen soon if I continue to play.

Edited by Duke Nedo, 03 September 2014 - 11:32 PM.

[Kyrie]

The problems in the match maker are fairly simple to explain, IMHO.

1. Too much complexity for the existing player-base (classes of mech, rating)
2. Leading to absurd relaxation of the parameters to get any kind of match going in a reasonable time.

The population is simply not large enough to support this complexity. We cannot be certain since PGI has hid the user-count ever since closed beta, but its likely that there simply are not enough players online seeking matches to make a complex MM work within a 60 second window.

While I hate to suggest extra complexity, there might be a way out of this conundrum by allowing users who really want to find a "good match" to set an "I'll wait!" parameter to avoid the relaxation of the MM protocol. This might have the effect of splitting the queues into two groups: those simply grinding out the content and do not really care, and those who want a solid playing experience.

Instead of trying to make a match within 60-120 seconds, if you (and others like you) are willing to wait up to say 10 minutes, the an equal match is much likelier/easier for the MM to obtain.

I humbly suggest that it is not likely the MM can be fixed as long as it is trying to juggle ELO as well as balancing a fixed set of mechs per tier within a short seek-time, but only PGI can know for sure by judging the actual user-counts and profiles.


EDIT: For clarity.

Edited by Kyrie, 04 September 2014 - 12:45 AM.

[J0anna]

I think one of the bigger problems is that not all wins or losses are the same. If the MM thinks you are going to lose and you do, your ELO doesn't drop very much. Same as if the MM thinks you should win and you do, your ELO doesn't rise very much. To get the proper readjustment to you ELO (i.e. to stop streaks) you need to get matches against what the MM thinks.

Also there appears no controls for the MM to try to balance out your w/l ratio, i.e. the MM could put you on teams where you're predicted to lose 10 times in a row. You lose, but your elo hardly budges. We should be able to see our ELO and how it changes. We should also have controls that adjust what we are willing to put up with - for example I would love a "no trial mech" option - the downside is I might have to wait an average of 5 minutes for a match. Or a option "no matches with elo below XXXX", but now you might have to wait even longer.

I think what everyone was seeing last weekend, was many new/weaker players flooding the server - and large elo discrepancies on teams as the MM was trying to balance ELO's.

[Almeras]

Sandpit, on 03 September 2014 - 07:52 AM, said:

You go on a nice winning streak. You eventually begin facing opponents of higher skill level. It's not the MM or Elo trying "even out" your W/L records, it's simply your Elo on a steady increase until eventually you're playing against opponents "better" than you skill wise.

it's like equating a college football player being drafted to the pros and realizing that the speed of the game has increased dramatically. It's not that the player has gotten "bad", it's that he is now playing with others at the same or higher skill level than he is at
on the entire team
as opposed to facing off against 1-2 superstars on the opposing team.

Not how it works, in reality it works like this;

You go on a nice winning streak. You eventually the MM will load you team with lower elo players. The MM or Elo isn't trying to group you in games of equal skill buckets it's simple as your Elo steady increases it gives you a handicap to produce more challenging games
until eventually you're carry more and more teammates. Its worth noting however your opponents may have a better arrogate ELO but chances are the same superstar+handicap set up (then good luck if your the uller and he's the madcat).

It's like equating a college football player having a few highshchool players seeded on his team because he's won too many games. It's not that the player has gotten "bad", it's that he is now has to carry harder. Now he's playing with a handicap where as before the he wasn't the more he upsets the prediction the bigger his handicap will be.

[Ryvucz]

MatchMaker currently has this formula...

-ERROR-

[Duke Nedo]

Almeras, on 04 September 2014 - 01:39 AM, said:

Not how it works, in reality it works like this;

You go on a nice winning streak. You eventually the MM will load you team with lower elo players. The MM or Elo isn't trying to group you in games of equal skill buckets it's simple as your Elo steady increases it gives you a handicap to produce more challenging games
until eventually you're carry more and more teammates. Its worth noting however your opponents may have a better arrogate ELO but chances are the same superstar+handicap set up (then good luck if your the uller and he's the madcat).

It's like equating a college football player having a few highshchool players seeded on his team because he's won too many games. It's not that the player has gotten "bad", it's that he is now has to carry harder. Now he's playing with a handicap where as before the he wasn't the more he upsets the prediction the bigger his handicap will be.

Is this official? Heard a few talk about it... if it's true it's actually good news because then all they need to do to fix it is to restrain the scaling of the handicap so that it isn't that aggressive.

[Black Ivan]

Match Maker - critical failure

[Sandpit]

Almeras, on 04 September 2014 - 01:39 AM, said:

Not how it works, in reality it works like this;

You go on a nice winning streak. You eventually the MM will load you team with lower elo players. The MM or Elo isn't trying to group you in games of equal skill buckets it's simple as your Elo steady increases it gives you a handicap to produce more challenging games
until eventually you're carry more and more teammates. Its worth noting however your opponents may have a better arrogate ELO but chances are the same superstar+handicap set up (then good luck if your the uller and he's the madcat).

It's like equating a college football player having a few highshchool players seeded on his team because he's won too many games. It's not that the player has gotten "bad", it's that he is now has to carry harder. Now he's playing with a handicap where as before the he wasn't the more he upsets the prediction the bigger his handicap will be.

you missed the part about that not being Elo but PGI's implementation of Elo. What you're talking about is how PGI chose to implement an Elo system. I also pointed out a few other factors that PGI has thrown into the mix that skews Elo as well.

Duke Nedo, on 04 September 2014 - 04:58 AM, said:

Is this official? Heard a few talk about it... if it's true it's actually good news because then all they need to do to fix it is to restrain the scaling of the handicap so that it isn't that aggressive.

No, it's not official, it's the observations of players. (unless they posted something on this that I missed)

[Bilbo]

Duke Nedo, on 04 September 2014 - 04:58 AM, said:

Is this official? Heard a few talk about it... if it's true it's actually good news because then all they need to do to fix it is to restrain the scaling of the handicap so that it isn't that aggressive.

No it isn't. The MM selects a target Elo and fills in as best it can. The acceptable range of Elo scores are widened as wait time increases. I guess it makes people feel better about themselves to think otherwise. This will, of course, mean that very high and low Elo players will occasionaly end up on teams they probably have no business being on.

[Sandpit]

Bilbo, on 04 September 2014 - 07:29 AM, said:

No it isn't. The MM selects a target Elo and fills in as best it can. The acceptable range of Elo scores are widened as wait time increases. I guess it makes people feel better about themselves to think otherwise. This will, of course, mean that very high and low Elo players will occasionaly end up on teams they probably have no business being on.

I think it's happening more often than occasionally at certain Elo ranges though. I see a dramatic difference in team composition when I drop in an assault as opposed to a light. (Assaults being my personal "high" Elo mechs and lights being the "low" Elo mechs)

[Almeras]

Duke Nedo, on 04 September 2014 - 04:58 AM, said:

Is this official? Heard a few talk about it... if it's true it's actually good news because then all they need to do to fix it is to restrain the scaling of the handicap so that it isn't that aggressive.

Its was mentions as an adjustment to the MM when the clans came out I believe it was Paul who said it.

[Sandpit]

Almeras, on 04 September 2014 - 08:19 AM, said:

Its was mentions as an adjustment to the MM when the clans came out I believe it was Paul who said it.

[citation needed]

[BigFatGator]

Would definitely like to know if there was a change to matchmaker in one of the last patches. The way I understood it, there should only be a certain range of Elo per team- there should not be a couple studs and 10 scubs as described by several in this threads and on the forums.

All I can think of is that perhaps the pool of high Elo players in the PuG que is so low that there is no chance for matchmaker to field two teams of high tier players without violating whatever clock the MM is trying to use for time to form matches. So then it takes a couple of the higher Elo players and cobbles together a team that it throws against a team with more even mid-level Elo.

To account for the streaks, all I can think of is that these times of low pools of Elo players are transient- so if a higher tier Elo player keeps trying to run a match during one of these times, they'll keep being in the situation where they have to carry a bunch of scrubs.

Would be great if PGI can confirm if this is what is happening, because then the fix is to switch to a class of mech that you don't do as well in during periods when there are low high-skill pools. Or figure out if there is a pattern to time of day/week when the streaks are occuring.