87 replies to this topic

[Ghogiel]

Mawai, on 25 September 2014 - 04:43 AM, said:

3) Correct. Except the maximum change is 20 not 50 and assigning the maximum should happen very rarely since the amount assigned depends on how much the prediction was wrong. If the team balance is 51:49 ... you might move 1 point.

Unless it's changed to 20 when it was rewriten a little while ago, it's still 50. Correct me if I am wrong with dev quote. (at least one thing has changed in this image, and thats the first time player elo, which gets a couple hundred points knocked off during cadet bonus)

Edited by Ghogiel, 25 September 2014 - 04:59 AM.

[Valore]

The only thing we can say for sure for Elo...

Is that its not an abbreviation. Its actually Elo.

The other thing, would be that no matter how good the Elo system is, YOU'RE always the one being screwed over, never the other team...

[Lily from animove]

is it even working. I am probably a bit above the average gamer.
But when I and my low elo friend drop in group queue vs Lords premades, I doubt Elo is even enabled.

[somanov]

(Apologies in advance if this has been answered in some other place)

Until now I had very little idea how ELO / matchmaking in this game actually works. Now that I read through the explanations in this thread I believe I have a little more clue and a question:

I am wondering if the mech a player is currently driving is somehow factored into matchmaking (beyond enforcing 3/3/3/3)

Simply put: If we have 2 players with the same (high) ELO score, one is sitting in a meta Direwolf, the other in a Locust. Would the matchmaker put the two on opposing teams and call it a "perfect match"?

[Jorgandr]

To abbreviate everything in this thread reagarding ELO:

After your cadet bonus is over, your ELO is set at "average"

Your ELO will only very rarely increase or decrease by 1 or 2 points (out of 2800), and is based on the score of the entire team (never just you).

Your ELO has nothing (or at least very very little) to do with your personal skill level. It is based on the teams that you roll with. So for the solo queue, it is random at best.

Conclusion - everyone in the solo queue is running at the exact same, or near the same, ELO level. ELO is thus rendered pointless in solo queue.


Little personal experience: I regularly see trial mechs in my games, so I can thus assume that I am in the "middle range" bracket. I also, however, still regularly see people that I know often play with set competitive teams (because I watch them on youtube/twitch), so one would imagine their ELO must be fairly high (higher than mine at least).... so which is it? Am I in the high range, or the mid-range?

Personally I think ELO is just plain broken.

somanov, on 25 September 2014 - 05:47 AM, said:

...

Simply put: If we have 2 players with the same (high) ELO score, one is sitting in a meta Direwolf, the other in a Locust. Would the matchmaker put the two on opposing teams and call it a "perfect match"?

I highly doubt matchmaker even makes a passing glance at individual mechs outside of "it is a light, or it is a heavy". Otherwise you would never see a match with zero ECM on one side, and 4 on the other in the solo queue.

I would just love for PGI to make your ELO value visible (at least to yourself). That way everyone can take a screenshot and we can see the truth: Is everyone pretty much at or near 1400? (I predict yes). I doubt they ever will though, since they publicly base the effectiveness of their balance decisions on their "perfect" ELO system.

Edited by Jorgandr, 25 September 2014 - 06:37 AM.

[somanov]

Jorgandr, on 25 September 2014 - 06:17 AM, said:

....

I highly doubt matchmaker even makes a passing glance at individual mechs outside of "it is a light, or it is a heavy". Otherwise you would never see a match with zero ECM on one side, and 4 on the other in the solo queue.

ok, but even when not looking at individual mechs. Would the matchmaker say two players are equally matched when the one is in an assault and the other in a light?

I am probably just a sub-average player so my personal experiences may not count for much. But in my experience many lights are just not very useful until the brawling starts and there are heavily damaged mechs to be critted. However a match is often already decided when the brawl starts, so even very good players in lights might never get the opportunity to play out the strengths of their mechs/builds until it is too late to change the outcome of the game..

(Disclaimer: Dont get me wrong I love my Firestarters, I just dont think they are overly useful until there are a few heavily wounded enemies around which I can try to backstab.)

[Jorgandr]

somanov, on 25 September 2014 - 06:36 AM, said:

ok, but even when not looking at individual mechs. Would the matchmaker say two players are equally matched when the one is in an assault and the other in a light?

...

I have to assume that matchmaker relies 100% on ELO values, and also attempts to make things 3/3/3/3 when possible. I have never seen or heard anything to indicate otherwise. The locust vs DW isnt quite a fair question though, as it is trying to match assaults to assaults and lights to lights (via the 3/3/3/3 system, which does not always work).

It would be more fair to have the system compare a LCT-1V (MG Locust) to an FS9-K (Laservomit). In that case, I suspect the answer is yes: Matchmaker considers them to be the same thing, even though a poorly piloted FS9-K could probably solo 4 pro LCT-1V's.

Edited by Jorgandr, 25 September 2014 - 06:45 AM.

[Bilbo]

Ghogiel, on 25 September 2014 - 04:57 AM, said:

Unless it's changed to 20 when it was rewriten a little while ago, it's still 50. Correct me if I am wrong with dev quote. (at least one thing has changed in this image, and thats the first time player elo, which gets a couple hundred points knocked off during cadet bonus)

It was changed. when it was rewritten. The Karl Berg thread, in off-topic, has much goodness in it and should be required reading before people start discussing the MM and Elo.

[Dremnon]

Honestly, it's time to get away from the ELO system altogether, it simply doesn't work. The fact that we are here a year later and there are still discussions related to how groups are assigned in solo que pugs shows that system doesn't work, or that how the system is supposed to work isn't being clearly explained by the Devs.

I'm also approaching 8000 matches lifetime, and through it all my questions has always been why can't groups be assigned by the number of matches played, either overall lifetime, or per weight class. It makes far more sense to have people grouped together that have played around 2000 +/- "x" matches in a certain weight class (i.e. 3/3/3/3 1750 to 2250 matches) than potentially having first time players drop with people that have played 8000 matches. New players should drop against new players or fairly new players (i.e. 0 to 250 matches lifetime, or 0 to 50 by weight). Have an incremental system that doesn't require a mathematical equation to figure out things. Keep it simple, logical, and fair. That way new players won't get discouraged when they get stomped their first 50 matches and want to quit.

~D

[Ghogiel]

Bilbo, on 25 September 2014 - 06:45 AM, said:

It was changed. when it was rewritten. The Karl Berg thread, in off-topic, has much goodness in it and should be required reading before people start discussing the MM and Elo.

You are going to have to link to the dev quote.

[Bilbo]

http://mwomercs.com/...ost__p__3710094

[Ghogiel]

Bilbo, on 25 September 2014 - 07:33 AM, said:

http://mwomercs.com/...ost__p__3710094

Cheers boss

[jackal40]

I still don't understand how my ELO does not rise or fall based on my performance in the match. If I do 0 damage (for whatever reason) in the battle, my ELO should not rise - I didn't do anything to contribute to the win. If I no more than half the total team damage, I should get a substantial rise in ELO.

Since the deck is already stacked against me by match maker, maybe we should get a cbill bonus when we loose as expected.

[Sjorpha]

Jorgandr, on 25 September 2014 - 06:44 AM, said:

I have to assume that matchmaker relies 100% on ELO values, and also attempts to make things 3/3/3/3 when possible. I have never seen or heard anything to indicate otherwise. The locust vs DW isnt quite a fair question though, as it is trying to match assaults to assaults and lights to lights (via the 3/3/3/3 system, which does not always work).

It would be more fair to have the system compare a LCT-1V (MG Locust) to an FS9-K (Laservomit). In that case, I suspect the answer is yes: Matchmaker considers them to be the same thing, even though a poorly piloted FS9-K could probably solo 4 pro LCT-1V's.

As I understand it you also have separate ELO values for each weight class, so you are seeded differently depending on how good you have played with that class.

Honestly I believe a lot tof the complaining is based in people overestimating their own skill, thinking they are so much better that the "bads" they are being matched with, when in reality they are mediocre to average players and simply seeded as such.

[Jorgandr]

Sjorpha, on 25 September 2014 - 08:34 AM, said:

As I understand it you also have separate ELO values for each weight class, so you are seeded differently depending on how good you have played with that class.

Honestly I believe a lot tof the complaining is based in people overestimating their own skill, thinking they are so much better that the "bads" they are being matched with, when in reality they are mediocre to average players and simply seeded as such.

But matchmaker has no clue how you, personally, perform. ELO is tracked based on whether your entire team full of random people won or lost.

Also, according to the devs themselves matchmaker almost never has large discrepancies between ELO numbers, so your ELO likely almost never changes appreciably no matter what you do.

I suspect I could pike every single match and have almost no change in my ELO overall.

Probably the only way you could decrease your ELO is to spend every match TK-ing... and even then it would probably take 1000 matches to go down a "tier"... depending on where their nebulous marker for such a thing is set (Does matchmaker consider a team with 11 people with an ELO score of 0 and one with 2800 vs a team of 12 with 233 each a "fair" match?) .

Edited by Jorgandr, 25 September 2014 - 09:17 AM.

[Bilbo]

Jorgandr, on 25 September 2014 - 09:01 AM, said:

But matchmaker has no clue how you, personally, performed. ELO is tracked based on whether your entire team full of random people won or lost.

Also, according to the devs themselves matchmaker almost never has large discrepancies between ELO numbers, so your ELO likely almost never changes appreciably no matter what you do.

I suspect I could pike every single match and have almost no change in my ELO overall.

It's not the discrepancy in Elo that determines the amount and direction of your change in Elo. It's the degree of confidence the MM has in the outcome. Go ahead and pike those matches, I'd be willing to bet it takes you longer to get out of the hole you just dug yourself than it did to get you there in the first place.

[jozkhan]

We dont actually 'know' anything about Elo in MWO because it is entirely hidden from us in terms of our own Elo, other players Elo and everything else in fact.

It could well be the biggest joke around the PGI studio and we've just been using a flat match maker all along.

We 'know' - as in can verify independently and confirm - absolutely nothing.

Sure we've been told alot but we cant verify any of it.

Edited by jozkhan, 25 September 2014 - 09:19 AM.

[Jorgandr]

Bilbo, on 25 September 2014 - 09:09 AM, said:

It's not the discrepancy in Elo that determines the amount and direction of your change in Elo. It's the degree of confidence the MM has in the outcome. Go ahead and pike those matches, I'd be willing to bet it takes you longer to get out of the hole you just dug yourself than it did to get you there in the first place.

But how accurate is the MM in predicting that outcome? The only way your ELO changes is if MM got it wrong... and if MM is incredibly good at getting it right... then it was only wrong because something random happened, probably nothing to do with you personally. A disconnect, an afker, not having any ECM on your team. Any one of these things is enough to skew the results one way or the other, especially if every MM match is as close as PGI claims.

There are so many variables that definitely are NOT tracked by MM, that I have zero faith in the validity of its prediction system.

As far as I'm concerned the MM prediction system is not likely to be significantly better than a shot in the dark. It probably gets it right slightly more than 50% of the time. Any given persons score probably goes up just as much as it goes down. Everyone in the game probably has the same ELO.

jozkhan, on 25 September 2014 - 09:17 AM, said:

We dont actually 'know' anything about Elo in MWO because it is entirely hidden from us in terms of our own Elo, other players Elo and everything else in fact.

It could well be the biggest joke around the PGI studio and we've just been using a flat match maker all along.

We 'know' - as in can verify independently and confirm - absolutely nothing.

Sure we've been told alot but we cant verify any of it.

My point exactly.

They need to stop trying to match something a computer cannot possibly know, and stick to matching things like loadouts, weight, and known specific mech effectiveness.

Edited by Jorgandr, 25 September 2014 - 09:36 AM.

[Bilbo]

Jorgandr, on 25 September 2014 - 09:26 AM, said:

But how accurate is the MM in predicting that outcome? The only way your ELO changes is if MM got it wrong... and if MM is incredibly good at getting it right... then it was only wrong because something random happened, probably nothing to do with you personally. A disconnect, an afker, not having any ECM on your team. Any one of these things is enough to skew the results one way or the other, especially if every MM match is as close as PGI claims.

There are so many variables that definitely are NOT tracked by MM, that I have zero faith in the validity of its prediction system.



My point exactly.

In a single match you are correct, but over time not so much.

[Jorgandr]

Bilbo, on 25 September 2014 - 09:33 AM, said:

In a single match you are correct, but over time not so much.

And only if MM is actually effective at predicting the outcome. If it is effective, then it would take tens of thousands of matches to see significant change.

If it is not effective, then you would never see significant change.

All a moot point though, since we'll never know either way.

Keep in mind I am talking about the solo queue.

Edited by Jorgandr, 25 September 2014 - 09:44 AM.