20 replies to this topic

[SuomiWarder]

We have not had any detailed info on how Elo has been implimented in the logic it uses, but the following occurred to me based on what we do know.

We all start with the same Elo per chassis and change is slowish, so we can assume that the range of Elo scores is a bell curve. Only a small number of pilots are on the very high or very low end.

After about 2 minutes Elo gives up and groups who ever is still in the system waiting.

It follows then that the majority of players will be within a standard deviation of the starting value and find matches quickly. Leaving only the two ends - high and low- waiting for a match.

If there is not a matching high or low end, then the two will get tossed together based on time.

Thus I wonder what if any match making adjustments have been made to keep the highest scores from often facing the lowest scores when two minutes approaches and there is not a corresponding value. (I am assuming that the people in the mid range have already been matched up and launched).

[iminbagdad]

I doubt this is how it works, possible but doubtful. It would explain some things though. ( I'm not very high but I group with guys who are.)

[Roughneck45]

I don't think it works like that.

More likely, either group on the extreme end of the curve would get matches against people in the middle, not on the opposite end.

[Mechteric]

SuomiWarder, on 28 February 2013 - 10:52 AM, said:

We have not had any detailed info on how Elo has been implimented in the logic it uses, but the following occurred to me based on what we do know.

break out your math helmet

http://mwomercs.com/...ost__p__1626065

the more you know

Edited by CapperDeluxe, 28 February 2013 - 11:06 AM.

[Mawai]

I doubt that is how it works. In addition, I think it would then really only become a problem if the pool of players was too small. I'd also like to add that I have never waited 2 minutes for a match yet since they re-installed the revised Matchmaker 3.0. The responses seem to be (a) almost immediate (<10s) and about 30s. I think the first reduction in limitations occurs at about 20s. At least that is my impression ... maybe we should record some timing statistics

As far as balanced goes ... I think the constraints are ELO and total tonnage at the moment. I have seen a lot of matches with 3 lights, 3 assaults, 2 heavy/medium vs 1 or 2 assaults and 6 or 7 heavy/medium. I don't think these ones are really fair since I think the assault+light teams are usually more effective.

It would be nice if PGI could write up the logic used in the matchmaker just for interest sake. I would think that many folks would like to know how it comes up with the teams.

The one aspect I am particularly curious about is how the matchmaker factors in the ELO of pre-made teams when combining them with other players. It is very difficult to tell at times when there are pre-mades in the games I have been playing.

[SuomiWarder]

I re-read the link posted (which I had read before). I failed to find anything that notes what this Elo system does when having a problem looking for a match. If it picks the next closest, then we have no issue with the best facing the worst. If it picks randomly from whatever is still waiting for a match then eventually we might have a problem. Maybe I missed where ever the info was stashed though.

At the moment, the "very best" and "very worst" scores are probbaly pretty close in value. The only solace that those in the lower end have if facing the higher end is that the system will not reward the higher end with an increased Elo for stomping them.

[Mechteric]

SuomiWarder, on 28 February 2013 - 11:11 AM, said:

I re-read the link posted (which I had read before). I failed to find anything that notes what this Elo system does when having a problem looking for a match. If it picks the next closest, then we have no issue with the best facing the worst.

Good point that linked post didn't really have that particular bit of the info, but it was mentioned somewhere in some other post I don't care to search for. Basically as I understand it you have it right, it tries to find the closest "percent chance to win" or "odds to win" as it can then if there's not enough players around that match those odds then it expands those odds out. So yeah there's a chance you could have a team thats worse or better by a significant margin, but I think that's only if you're sitting there at the Searching screen for like a minute or more. Or if its new players who haven't yet made it to their appropriate Elo range (be it high or low since everyone starts in middle of the field)

Edited by CapperDeluxe, 28 February 2013 - 11:20 AM.

[Bilbo]

SuomiWarder, on 28 February 2013 - 11:11 AM, said:

I re-read the link posted (which I had read before). I failed to find anything that notes what this Elo system does when having a problem looking for a match. If it picks the next closest, then we have no issue with the best facing the worst. If it picks randomly from whatever is still waiting for a match then eventually we might have a problem. Maybe I missed where ever the info was stashed though.

At the moment, the "very best" and "very worst" scores are probbaly pretty close in value. The only solace that those in the lower end have if facing the higher end is that the system will not reward the higher end with an increased Elo for stomping them.

The system widens the band of acceptable opponents as time passes. The longer the wait the farther apart the Elos can be, as well as weight classes.

From my experience, the very best and very worst scores are nowhere near each other. The class of teammate and opponents I recieve when playing my light mechs solo is vastly different than those I recieve when playing my Atlas.

p.s. I suck at piloting lights.

[Star Captain Obvious Kerensky]

Because the Elo system widens the matching threshhold as the search runs, the Elo system dilutes its own usefulness. One team will have an advantage over the other; by design to prevent the matchmaker from failing to find a match.

On a long enough timeline with this behavior, the Elo system will be moot since the scores have been polluted by too many uneven matches.

Edited by Eldragon, 28 February 2013 - 11:28 AM.

[Roughneck45]

Eldragon, on 28 February 2013 - 11:27 AM, said:

On a long enough timeline with this behavior, the Elo system will be moot since the scores have been polluted by too many uneven matches.

On a long enough timline, it will get better. This is because even though it may be matching high ELO against low ELO the change in ELO is what you are really looking for after the match. Those skilled players are not going to get much for a win, because they are expected to win. It is the opposite for the the lower ELO. If they win, they will recieve a higher ELO gain from beating someone who was supposed to be better than them.

The toughest time for ELO is when it is first implemented, simply because it does not have enough data to accuratley assess what a players ELO should be.

Its not perfect, but I think it helps quite a bit.

Edited by Roughneck45, 28 February 2013 - 11:34 AM.

[Belorion]

SuomiWarder, on 28 February 2013 - 10:52 AM, said:

After about 2 minutes Elo gives up and groups who ever is still in the system waiting.

I think they way they phrased it, was that the rules are relaxed. That isn't the same as throwing elo out the window. I doubt that at that point it just takes the two that have been in queue the longest. The logical thing to do would be to start increasing the range until a match is made. Since both the highest and lowest would have the bulk of the bell curve to traverse before being matched up against the other extreme, it would therefor be very unlikely that the highest on the curve would ever be matched with the lowest.

[Vasces Diablo]

No. It starts with a target number and tries to match players with that number.

Also, just the very nature of a bell curve (assuming that's what the population looks like) has very few data points at the extremes. The vast majority of players (like 85% if I remember right) are all within one standard deviation.

[Jonathan Paine]

One slight problem with the Elo system right now is putting together a team. At times there can be a wide range between the highest rated player on a team, and the lowest rated player on a team. This generates a disparity where you sometimes see matches with the "supremo Elo person" getting 5+ kills, while one or more players spend the match firing lrms into the ground and bumping into buildings.

Why is this a problem? From what I have seen in matches both the top and the bottom often end up wanting to rage quit, and this does not foster a happy in-game atmosphere.

[Star Captain Obvious Kerensky]

Roughneck45, on 28 February 2013 - 11:30 AM, said:

On a long enough timline, it will get better. This is because even though it may be matching high ELO against low ELO the change in ELO is what you are really looking for after the match. Those skilled players are not going to get much for a win, because they are expected to win. It is the opposite for the the lower ELO. If they win, they will recieve a higher ELO gain from beating someone who was supposed to be better than them.

Since we are not privy to the exact calculations used, that may be true. However I am exceptionally skeptical; and I take your assessment as being overly optimistic.

You must remember that teams are a mix of ELO scores. If two skilled players are mixed with two new players and lose against four average players, how does the Elo gain/loss affect the elite player versus the new player? This gets further compounded when there is a tonnage mismatch.

Elo, as originally devised works for 1v1 games, or teams with a consistent team roster.

This is a team game with random players and variable equipment loadouts, where the Elo values are averaged across the team. PGI's implementation needs to account for all the variables or the Elo system would be polluted.

To compare it to chess...

This is 8v8 chess, where each player has their own Elo score and controls two pieces, and sometimes one team has more queens and knights than the other. Furthermore half of the players on one team cannot talk to each other.

If someone could show me an Elo scoring system for that hairball that works, then I will be less skeptical of MWO's system.

Edited by Eldragon, 28 February 2013 - 11:59 AM.

[urmamasllama]

it doesn't give up it widens it's search further out. you won't get the very best with the very worst unless none of the average players are queueing

[Madgunmurphey]

I think it'd be nice to get a developers stand point on this, because it appears to make or break the game. In the matches I play (usually in a 4 man lance) I would say 40% of the time we end up defeating the team 8-0 or 8-1 (what feels to be a blow out). I would say 40% of the time I end up being rolled over (totally annihilated). Its seems at about 10% percent of the time there is a great match that comes down to 1 or 2 kills. 10% of the time people are lazy and afraid to play and would rather win by a cap. So at a stand point 50% of the time players are satisfied and the other 50% of the time players hate the game.

[Roughneck45]

Eldragon, on 28 February 2013 - 11:54 AM, said:

If someone could show me an Elo scoring system for that hairball that works, then I will be less skeptical of MWO's system.

Yeah, it would be nice to know for sure.

I feel like you have to think that PGI would have at least considered the situation though. I would guess that the higher ELO player is still walking away with either a smaller ELO gain, or a larger ELO loss, for a match when they are by far the highest ELO.

So, my guess, is that each player recieve/loses their own amount of ELO, rather than the team all gaining the same amount or losing the same amount.

On a side note, all of my matches have been better since ELO came out, so it seems to be working in my eyes.

EDIT: Hmmm, so after reading that link a bit more thoroughly, it does say that it takes average ELO for the team and uses that for the gain/loss. Not quite sure what to make of it now.

Edited by Roughneck45, 28 February 2013 - 12:30 PM.

[Bilbo]

Eldragon, on 28 February 2013 - 11:54 AM, said:

Since we are not privy to the exact calculations used, that may be true. However I am exceptionally skeptical; and I take your assessment as being overly optimistic.

You must remember that teams are a mix of ELO scores. If two skilled players are mixed with two new players and lose against four average players, how does the Elo gain/loss affect the elite player versus the new player? This gets further compounded when there is a tonnage mismatch.

Elo, as originally devised works for 1v1 games, or teams with a consistent team roster.

This is a team game with random players and variable equipment loadouts, where the Elo values are averaged across the team. PGI's implementation needs to account for all the variables or the Elo system would be polluted.

To compare it to chess...

This is 8v8 chess, where each player has their own Elo score and controls two pieces, and sometimes one team has more queens and knights than the other. Furthermore half of the players on one team cannot talk to each other.

If someone could show me an Elo scoring system for that hairball that works, then I will be less skeptical of MWO's system.

The teams Elo score is averaged. The difference between the two team's average determines the amount of change for each player's Elo in either direction. The most you can gain or lose in amy match is 50. The closer you get to an exact match of the team's average Elo, the smaller the change to everyone's Elo.

[Star Captain Obvious Kerensky]

Bilbo, on 28 February 2013 - 12:15 PM, said:

The teams Elo score is averaged. The difference between the two team's average determines the amount of change for each player's Elo in either direction. The most you can gain or lose in amy match is 50. The closer you get to an exact match of the team's average Elo, the smaller the change to everyone's Elo.

That does not answer if the Elo system is poluting itself. If you have a grossly mismatch game, but the ELO system decides the Average Elo is the same, the resulting Elo adjustment is not reflective of actual player skill. Does the Elo system account for tonnage? Intra-team disparity in Elo score? 4-man vs. solo? Maybe it does, but PGI has been silent on the issue.

Seems to me PGI should have done this in baby steps. Have a "Elo Bar" rather than the current matchmaker. Players below the Elo Bar are in the "newbie pool", players above the bar are in the "Veteran Pool". Players in the two pools don't interact. Elo is only used for determining if you are above the bar. It prevents newbie-stomping as the current system does, but also prevents strangely mis-matched teams. After several MONTHS of the bar system, then maybe move to an individualized system.

Edited by Eldragon, 28 February 2013 - 12:27 PM.

[Bfvmg]

SuomiWarder, on 28 February 2013 - 10:52 AM, said:

We have not had any detailed info on how Elo has been implimented in the logic it uses, but the following occurred to me based on what we do know.

We all start with the same Elo per chassis and change is slowish, so we can assume that the range of Elo scores is a bell curve. Only a small number of pilots are on the very high or very low end.

After about 2 minutes Elo gives up and groups who ever is still in the system waiting.

It follows then that the majority of players will be within a standard deviation of the starting value and find matches quickly. Leaving only the two ends - high and low- waiting for a match.

If there is not a matching high or low end, then the two will get tossed together based on time.

Thus I wonder what if any match making adjustments have been made to keep the highest scores from often facing the lowest scores when two minutes approaches and there is not a corresponding value. (I am assuming that the people in the mid range have already been matched up and launched).

Lots of talk about timing statistics, etc.
One thing I have noticed is that the faster I get matched up, the more chance I have of a good match or being stomped...
And I get A LOT of matches that take a long time to match up, and the longer it takes, usually (note that qualifier there, usually) its me stomping the other team.

Lets face it. ELO has been out for a couple of weeks now, and at the rate most of us drop...there has to be a wide variation by now. Especially this last marathon weekend of drops and ELO adjustments.

So the longer it takes to match, it seems the more disparity there is in "Skill" level.
So I would say that Yeah, it seems to me anyway that is the way it works. At least for now.