233 replies to this topic

[Kaiser Thermidor]

Yes Piranha, let's talk about Elo, and "skill based" matchmaking in general.

Firstly, I have issues with using Elo in this game. Let me tell you why!

The Elo ranking system was originally designed for chess tournaments, with a numerical ranking that less gauged a players skill and more gauged their chances of winning or losing (It is likely a matter of opinion if these two things are actually seperate values). In a game such as chess, in which nearly every faucet of the game is under a single player's control, a win/loss chance is probably a good indication of their skill. If they lose, they only have themselves to blame.

However, last I checked, MechWarrior Online is a team game. We do not succeed or fail on our own abilities or merits. We have 7 other people (ideally, when the matchmaking doesn't throw a fit) on which to rely, and it is as much our teammates' decisions and 'skill' that will determine the outcome of any given match.
Additionally, MechWarrior Online is not a game of 1v1 duels; team work factors in to a very large degree, as you designed it to. Saying Team A and Team B each have 'an equal chance to win, roughly' doesn't mean much when it is entirely possible for players to be carried through games and thusly have their Elo ranking artifically boosted. If I have one particularly silly player who runs into the middle of an enemy lance and explodes in a several million Cbill fireball, it is quite possible my team can still win the match, and since Elo only cares about a win or a loss, that particularly silly player's Elo rank goes up (either a lot or a little, depending if we were the underdogs or not).

Now, to my understanding, the introduction of Elo was meant to give each team a fair shot at winning (though I don't think that was not the case before). If this is true, everyone should have a 1.0 or close win/loss ratio, as the law of averages should, if all odds are fair, bring us to a 1:1 ratio for *all* players that use the standard public matchmaking.
However, as of this writing, my win/loss ratio for conquest (the only mode I play) is 1.73. I'm not a particularly good player. This tells me that this implimentation of Elo is probably not working as intended.
But! This is a thread about an annoucement about changes to Elo, which implicitly states "Elo is borked." So maybe I'm right, about it not functioning as intended. To me, though, this less indicates 'we should fix our Elo system' and more 'Elo is needless and needlessly complicates the game'.

Now, I'll be completely up front about this: this is all based on my own understanding of Elo. From your Command Chair post about it, and from the Wikipedia article, that is as far as my knowledge of Elo rankings extends. I'm am neither a tournament chess player nor a statistician.

Additionally, what really bugs me, PGI, you already have a system for measuring 'skill' in this game: experience point rewards. Why not just use that? It's a mostly decent system that rewards teamplay and individual 'skill/ability'. Values would need to be tweaked, but nonetheless, it is vastly superior to a system that is designed for individual competition.

I used the latest Ask the Devs 36 to ask about Elo, but as it is only Monday, I've yet to see if my question is even worth an answer.

Also, yes, the weight matching has needed work for a long time now.
And those are just the more egregious that I remembered to screencap. It's endemic.

Anyway, my two cents. Hope you can take something useful away from it.

Edit: Typo

Edited by Vadim Krasvanya, 15 April 2013 - 01:48 PM.

[Eternal Hunter]

Matchmaking should IMO do the following.

Primary: Balance skills on both teams.
Secondary: Try to balance total team tonnage, OR assign a point value to mechs, items and equipment and match that roughly with each team.
Thirdly: Keep ECM mechs +/-1 and DO NOT EVER match a ecm vs no ecm team (cough, okay, i'm a bit testy with ecm, heh)
Fourth: Try to keep even classes between mechs if possible.

But we'll always have what might feel like unbalanced games.
Sometimes the other team just get into a bad position and the other team is in a great position.
Sometimes the other team steamrolls with the entire team and keeps on mowing the other one down while they all panic.
Someone might even get lucky and snipe a cockpit at the start of the match, creating low morale.
Heck, once i sneezed as i fired an alpha strike, missing completely, got overheated and died

The point is, every game is actually unique with different variables each time, so it's never gonna be "equal", but they can make it a lot closer, and that's what their planning as far as i read.

[Victor Morson]

Sprouticus, on 15 April 2013 - 05:15 AM, said:

It sounds like the total tonnage is fairly close in most cases (20-30 tons). Now it could be argued that the reason tonnage has been so close is that people are taking the hgh end of weight classes, but wathcing the number of Dragons, Catapults, Stalkers, etc in drops I dont think that is the case.

I would of course prefer tonnage based balancing (or even better ROLE based balancing), but that's just me.

In a serious game where it's class, not ton, you'll almost always take the same things. That "small drop" as you put it from 30 - 20 is enough to bring an Awesome to a Highlander or a Dragon to a Cataphract.

I don't get why they don't focus on tonnage matchmaking. Just scan the people actively looking to connect's tonnages and throw them into a game that'll put them within 50-60 tons of each other. It'd be way way way better than ELO even.

[codynyc]

yeah im not holding my breath... im surprised they are getting around to something other then trying to push OVERPRICED useless add on down our throats

[Dan Nashe]

I have been a big advocate of tonnage matching over class matching, admittedly.

But I'm starting to take a more "meh" approach. There's something to be said about the fact that
Stalker = Atlas = Highlander
Catapault=Jaeger=Cataphract
Trebuchet=Hunchback=Centurian=ECMCicada
(interestingly, you play a cicada, you get matched with other mediums, potentially giving your team a "light" advantage. This is interesting to me and encourages me to play a cicada 3M)

Light, Raven3L=Any Jenner

I think the Commando and Flea will always be lark mechs, and to a lesser extend the spider. Along with the
Non-ECM Ravens, Blackjack, non-ECM Cicadas (except maybe x5?) and Awesomes. But if you really love 'em, your elo should drop until they're playable again (so you get to kill atlases that are worse players).

And there's something to be said for medium + heavy =/= light + assault (the dark side of tonnage matching).

So I guess I've gone from a tonnage match proponent to sitting on the fence.

Overall, I like the changes proposed as better than the current set-up. So I'll call it a win.

I stand by my position that all non-Jenner lights should get ECM, and the Raven 3L should be given a 245-255 engine cap though. [Jenners retain their position as the combat-light but lack ECM. Other lights can now all perform the scout role (which requires ECM), and by being faster than the 3L, other lights won't be useless by comparison.]

IT probably affects my view that mediums all feel pretty bad right now, apart from tonnage.
And it feels like mechs have settled into:
Lights: Fast, hard to hit, XL
Heavies: Firepower, XL engines
Assaults: Standard Engines, hit takers

Mediums tend to be standard engines, like little assaults, but without enough speed for it to really help.

Edited by DanNashe, 15 April 2013 - 05:28 PM.

[Sam Slade]

DanNashe, on 15 April 2013 - 05:25 PM, said:

And there's something to be said for medium + heavy =/= light + assault (the dark side of tonnage matching).

So I guess I've gone from a tonnage match proponent to sitting on the fence..

with this point you've also put me in the fence sitting camp

[ShadowSpirit]

All this talk about matchmaking and ELO ... as it relates to weight and mechs.

1) What about map selection which should directly impact mech selection?

2) Why must every drop be 8 vs. 8? Why is it so limited? Why can't you do drop types and limit them by weight? Why can't pilots and groups have more control over the drop conditions?

3) Why are you so focused on ELO? Is the game *always* going to be "click ready" and get a "random match" ? I sure hope not.

... there is so much untapped potential in this game. It seems like we are missing the forest for the trees here.

I don't know why you are forcing everyone to be average with ELO. If this works out we're looking at a 50/50 win ratio? How is that even remotely fun? The universe is hostile ... why are you trying to tame it?

This is a shooter / simulation. Getting your @#@#$ stomped is motivation to get better. It seems you are trying to appease the masses and just really alienating the hard core gamers.

Edited by ShadowSpirit, 15 April 2013 - 06:48 PM.

[ShadowSpirit]

Followup: By the way ECM still sucks. You can't compare apples (DDC) and oranges (virtually any other assault).

[Deathlike]

Something to think about:

During MM v2 (no ELO, just tonnage matching), I intentionally took a Cicada (which was what I was grinding at the time after the 3L) for ECM and being a medium. The hunchback and centurion at the time (no trebuchet) was not a threat to me (although, the Cents have grown on me). The Cicada is an "overweight light" by many standards, and for the purposes of class matching, it screwed things up a little.

The same could be said for the Dragon.. it is an "overweight Medium" (at least to me). It simply does not compete in the same way as a Catapult or Cataphract at the time (v2 had no Jagermech). The Dragon is useful, but by class weight standards, it's not used in the same capacity and/or role.

The Awesome is another "overweight Heavy". It simply isn't in the same class as the other assault mechs (including the upcoming Highlander). The Awesome is still powerful for those that use them, but the role of the Awesome is very much closer to a Heavy than it is for an Assault mech.

Anyways... you have to really think about role warfare a little bit when it comes to MM and class weight matching... because there is a lot more impact than people think. A Cicada 3M+Orion (whenever the Orion is released) may be more similar to a Raven 3L+Awesome... but not all combinations are the same with class weight matching. Tonnage matching however is a lot more complex and requires a lot more thought to implement. I hope new mechs factor into some of the missing tonnages in order to make the "choices" better for a potentially better MM design.

[Zylo]

Roland, on 15 April 2013 - 05:16 AM, said:

But you are missing a key point here.

The Awesome isn't supposed to be equivalent to an Atlas. It's 15 tons lighter. It's supposed to be weaker.

The fact that it's 15 tons lighter is supposed to be the balancing point. Because it's supposed to exist in a game that is balanced on either tonnage or BV.

I'm not saying the Awesome should be able to stand in front of an Atlas while both shoot each other and expect to survive. The Awesome should be better in some areas to make up for the lower tonnage. Maybe it would be faster and more maneuverable for example, like the fast Awesomes in closed beta.

[Zylo]

Zyllos, on 15 April 2013 - 07:57 AM, said:

See, I beg to differ on that subject.

The Awesomes/Dragons dictate the fight against the Atlases. Even if the Cataphract/Atlas lance was together, the Awesomes/Dragons should be able to single out a Cataphract/Atlas target while only the Cataphracts could against the Awesomes/Dragons, either by superior speed (Dragons) or support range (Awesomes).

And you mentioned the balance been the RVN-3Ls and the Mediums. The Mediums should win the fight but the RVN-3Ls would be dictating the engagements.

I don't think you can dismiss a tonnage based balanced over the weight class balance, or vise versa.

We have tried the weight class based balancing, and we got what you would expect. Many mechs get left to the wayside, as the same happens with 100% free choice. The one choice we have not tried is tonnage based matching.

But it seems the devs are afraid to try this. Either in that it will take too long to match by tonnage, or they will force Timmy or Sally to have to pick a mech that he/she doesn't want to play.

The problem is that tonnage matching wouldn't solve the issue with less frequently played mechs in the small-group/solo random games. Players don't know which mechs will be used by the rest of their team so they generally take the best mechs if they don't want to put their team at a disadvantage. Tonnage matching would not change this practice at all in small-group/solo drops.

Weight class matching makes the most sense for these small-group/solo matches as players only have control over half the team at most (excluding sync-drop cheaters of course). When the matchmaking system tries to put together a team, it should be far easier to balance a team with small pre-made groups if the system only needs to consider 4 weight class possibilities along with Elo ratings. Since group sizes will vary between 2 and 4 players and solo players will be included the weight class matching results in fewer choices the matchmaker must consider.


In 8-man drops a min/max tonnage range could work better because each group would have control over each mech on their team. This might encourage a team to run lighter assaults to have tonnage available for an additional heavy mech rather than a medium for example.

[Denno]

I just want our n00bs to have fun and have a moderate chance of success against other n00bs. At least for a while. However they go about it, they need to get that right. Opinions here on how to get that done seems to vary wildly (as usual).

Dropped into River yesterday, for instance, and I encountered a trial Raven trying to cap. Too hot, slow, weak, too lightly armored and I doubted he even saw me trudging up. I was whaling on him before I even really knew exactly what he was. He lasted all of 3 seconds, then left. I felt a little bad for him.

Don't get me wrong. I am somewhat competitive and will try to rise to the occaision against other competetive folks, but I've made my bank and got my desired mechs and modules been around long enough to shrug off an ownage or team roll. A brand new guy...We want those folks to stay. You just aint going to get that with putting them in with vets. Dunno if tomorrow or Friday's changes is goign to do that.

[pesco]

Omid Kiarostami, on 12 April 2013 - 11:44 AM, said:

To clarify, it'll be similar to the forced Weight Class balancing that we used to have. We're not doing any tonnage based matching at this time.

Why not? It has been the standard for multiplayer Mechwarrior matches forever and doesn't have the obvious disadvantage pointed out in the question you replied to.

[Black Ivan]

Matchmaking should be by tonnage, not weight class.

[ShadowDarter]

this is the first version of this add-on, let the vultures pick over the remains of the new data in the next few weeks and work out the next step, remember things will take time.

[Zylo]

Black Ivan, on 16 April 2013 - 03:56 AM, said:

Matchmaking should be by tonnage, not weight class.

Tonnage matching has far more potential matchmaking issues when pre-made groups are considered.

Weight class matching allows matchmaker to easily match up multiple pre-made groups per team because only 4 possibilities for weight class per player must be considered rather than 12 different mech weights currently. The number of mech weights will increase over time as new mechs are added.

Tonnage matching could work for 8-man groups but the solo/small-group matches need a simple system like weight class matching to deal with the groups ranging from 2 to 4 players combined with lone-wolf players. Once Elo ratings are considered even weight class matching may be difficult during off-peak hours.

Edited by Zylo, 16 April 2013 - 04:27 AM.

[Divine Madcat]

Black Ivan, on 16 April 2013 - 03:56 AM, said:

Matchmaking should be by tonnage, not weight class.

Agreed. Nothing like taking your favorite Stalker mech, and knowing you just gave the enemy an Atlas.
Or when you take a Commando, knowing the team probably got a Raven to counter you.

Grouping by overall tonnage is the only smart and fair way to go.

[Roland]

Tonnage matching makes the most sense when you have a lobby system.

However, in totally randomized matching, I understand the desire to use weight classes.

It sucks to take an awesome and know that you're gonna be matched against an atlas, but it's potentially better than ending up on a team with 8 hunchbacks, and getting matched against 2 Atlases and 6 lights.

[Zweistein]

Question here: Is the framerate drop may of us experienced with the last patch (not hotfix) going to be improved? I lost 10 FPS that day which means the world for me since my original FPS was 25. Hotfix only have me like 3 FPS back.

[Sprouticus]

Divine Madcat, on 16 April 2013 - 05:44 AM, said:

Agreed. Nothing like taking your favorite Stalker mech, and knowing you just gave the enemy an Atlas.
Or when you take a Commando, knowing the team probably got a Raven to counter you.

Grouping by overall tonnage is the only smart and fair way to go.

using the same logic, it should only match chassis for chassis or even variant for variant. PGI has to balance how flexible the matchmaker is against the mechs dropped. If you make it more specific on mechs, then the balance is to either wait longer, or expand the ELO range out so you end up with better/worse players in the right mech.


Not to mention this is not how they do matchmaking (fitting the mech in the slot. The match against ELO 1st, then other stuff (if I understand it correctly.