Jump to content

Elo Is For Chess, Not Mwo


198 replies to this topic

#101 Killstorm999999

    Member

  • PipPipPipPipPip
  • 196 posts

Posted 09 October 2014 - 08:20 AM

View PostGhogiel, on 09 October 2014 - 08:15 AM, said:

So is PGI. That's why they put exactly that in the game.


They did not, they put in private matches that do not earn rewards.

A lobby system is a list of games that people made. Players can join one they find interesting. Once there are enough players and everyone is ready it can be started. Once the game is complete, it can rollover into another games. Perhaps the players can vote for a new map, or the original creator can just pick one. Players would be able to leave and new ones can join. Most importantly, this would allow you to build a consistent relationship with players over the course of a few games. If it turns out the other players are too good, or not good enough, you can drop out and join a different one.

#102 1453 R

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Bridesmaid
  • Bridesmaid
  • 5,569 posts

Posted 09 October 2014 - 08:22 AM

People.

For the eight millionth time.

"Battle Value" is a massively complex, impossible to balance, unacceptably finnicky system that is just as prone to mismatches, min-maxing, and gaming-of-the-system as MWO is right this second. It is not a magic band-aid that can be coded in a day, tested in another day, and shipped out to live on the third day and from there proceed to solve every single balance problem MWO has ever had forevermore.

This isn't even touching on how time-consuming and resource-intensive such a system would be to implement, or how doing some manner of convoluted BV system would toss game balance as ti stands completely out the airlock, or - and here's the real killer - how some equipment performs more effectively on certain 'Mechs or in certain builds than it does in others.

Super-easy counterpoint for you - a stock Nova-Prime gets X Battle Value. Now, the exact same 'Mech, except with two C-ERML removed from each arm, leaving a total of eight medium lasers, would get X - (C-ERML BV *4), thus showing that the modified NVA-Prime is less effective than the stock model. Except it's not. Even leaving those four tons empty, as opposed to stuffing them with heat sinks the way most folks do, would yield much improved heat efficiency and a distinct lessening of the ability to kill oneself with Ghost Heat, allowing pilots to be more consistent with the 'Mech and put up better numbers, leading to an improvement in performance rather than the predicted drop in performance that BV expects.

That is one example off the top of my head, pulled out of thin air in less than two minutes by a total casual. And you guys truly, honestly expecting that ultracomp league-playing super-mega-hardcore death-before-defeat players aren't going to find a million and one ways to break BV wide open, even after Piranha stops development of other features for three months to try and assign its own specific little predictable numeric value to each and every single possible player choice in the game?

#103 Fire and Salt

    Member

  • PipPipPipPipPipPipPip
  • 526 posts
  • LocationFlorida

Posted 09 October 2014 - 08:27 AM

ELO is great for veterans.

Stomping a bunch if scrubs isn't as fun as getting pitted against other to level players.

Speaking as a higher ELO player, I am glad that I get challenging matches.
How am I supposed to get better if I only play against people below my skill level?




It just needs to GTFO when tourney time comes around because it unfairly penalizes good players.

#104 IceSerpent

    Member

  • PipPipPipPipPipPipPipPipPip
  • 3,044 posts

Posted 09 October 2014 - 08:30 AM

View Post1453 R, on 09 October 2014 - 08:22 AM, said:

Super-easy counterpoint for you - a stock Nova-Prime gets X Battle Value. Now, the exact same 'Mech, except with two C-ERML removed from each arm, leaving a total of eight medium lasers, would get X - (C-ERML BV *4), thus showing that the modified NVA-Prime is less effective than the stock model. Except it's not. Even leaving those four tons empty, as opposed to stuffing them with heat sinks the way most folks do, would yield much improved heat efficiency and a distinct lessening of the ability to kill oneself with Ghost Heat, allowing pilots to be more consistent with the 'Mech and put up better numbers, leading to an improvement in performance rather than the predicted drop in performance that BV expects.


This is simply wrong - that Nova would get BV decrease from losing 4 MLs, but it would also get BV increase from getting better heat efficiency. Whether the net BV change is lower or higher depends on BV scheme (i.e. on how we "weigh" heat efficiency vs. weapons).

#105 Artgathan

    Member

  • PipPipPipPipPipPipPipPip
  • Knight Errant
  • Knight Errant
  • 1,764 posts

Posted 09 October 2014 - 08:32 AM

As always on this topic, we see the community segment into two groups: those who hate Elo and think it's impossible to use, and those that understand statistics and probability.

Is Elo the best possible matchmaker for MW:O? Probably not. Does it work? Yes.

#106 Alwrathandabout42ninjas

    Member

  • PipPipPipPipPipPipPipPip
  • The Formidable
  • The Formidable
  • 1,098 posts

Posted 09 October 2014 - 08:35 AM

View PostDiablobo, on 08 October 2014 - 08:40 PM, said:

You guys do understand how Elo works don't you? It takes two players who are assumed to have the EXACT SAME STARTING CONDITIONS AND CAPABILITIES, and then makes predictions on who should win when player X competes against player Z.

Have you figured out the flaw with using Elo for MWO yet? It's not too hard to see....if you don't see it, then you have no business even trying to comprehend the intricacies of Elo in the first place.

Elo assumes both players have the same basic starting point. Apart from white getting the first move, chess is an extremely balanced game. Both players have the same pieces, and both players' pieces can do the exact same things.

Is that what happens in MWO? Does someone who takes a zero unlock Commando with 3 small lasers, a standard engine, and single heat sinks stack up against a fully mastered, fully optimized Spider? No. They do not. Neither do the other mechs that come into the matchmaker that are not as powerful as the fully optimized meta builds. Using Elo with a game that has such widely divergent starting positions is just a recipe for disaster. It turns out that PGI has cooked up a wonderful dish of one-sided mismatch stew.

Until PGI ditches Elo and uses some sort of Battle Value system, there can be no matchmaker that is anything other than random, luck of the draw crap. The sooner they acknowledge this fact, the sooner we can all get on to enjoying our epic well matched battles we were promised. Instead, we have to be saddled by noobs on our team who think it is fun to pilot an Atlas with a couple of med lasers and some LRMs with single heat sinks and no unlocks, while the other team has fully unlocked and optimized meta builds and pilots who know how to use them.

There will never be a decent matchmaker as long as we use Elo. It is supposed to be used for teams that have identical starting positions. As we all know, that is not the case with MWO.


Well said. You have shared my feeling with the community with alot more detail. I have always beleived elo was meaningless in this game, and it still is. There are far too many variables for it to work in this game like you said.

#107 1453 R

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Bridesmaid
  • Bridesmaid
  • 5,569 posts

Posted 09 October 2014 - 08:50 AM

View PostIceSerpent, on 09 October 2014 - 08:30 AM, said:


This is simply wrong - that Nova would get BV decrease from losing 4 MLs, but it would also get BV increase from getting better heat efficiency. Whether the net BV change is lower or higher depends on BV scheme (i.e. on how we "weigh" heat efficiency vs. weapons).


Which only goes to illustrate my point. You honestly expect Piranha to accurately identify and program a correct Battle Value scale for each 0.01% of heat efficiency a 'Mech gains?

Assigning a BV point percentage to every last single tiny detail that can affect the outcome of a match is impossible. It. Is. Impossible. It is not a thing that is possible for sane human beings to accomplish. IT CAN'T BE DONE.

The only even remotely viable way for some sort of BV system to work is if it tracks usage. Usage, and nothing but usage. Player usage is a good barometer (pretty much the only good barometer) for generalizing and aggregating all of the innumerable, untrackable factors that contribute to a 'Mech's goodness or badness, and would be an effective counterweight to Elo if put to proper use. Even then however, a bad fit in a Timber Wolf would be weighed exactly the same as a good fit in a Timber Wolf, because it's impossible to program a system that accurately reflects every single one of the million and one minutae that go into 'Mech design and pilot aptitudes and preferences in order to create an Ultimate Goodness Quotient one can confidently saddle with 100% of the game's balance needs.

#108 KraftySOT

    Member

  • PipPipPipPipPipPipPipPipPip
  • The 1 Percent
  • 3,617 posts

Posted 09 October 2014 - 08:51 AM

Yup BV over ELO

#109 Gauvan

    Member

  • PipPipPipPipPipPip
  • 338 posts

Posted 09 October 2014 - 09:15 AM

Sorry, but ELO is the Electric Light Orchestra and is not widely used as a ranking system. Elo is the guy's name who developed the Elo rating system. :)

Having read through the thread, I think there are a lot of folks who do not understand how Elo works. Jon Phoenix linked to an explanation and I think reading that is important. Your rank is not negatively or positively impacted when you are matched with players well outside your skill level.

View PostJon Phoenix, on 09 October 2014 - 07:48 AM, said:

Is this still how Elo is calculated?

If you go in your profile and look at your win/loss ratio, and it is in the ballpark of 50/50, then Elo is working as intended to your benefit. That's data PGI has easy access to for the whole player base so if a majority of folks had W/L numbers outside a rage PGI considers acceptable they would change the MM algorithm. My W/L is near enough to 50/50 and if others are in a different situation that would be interesting to hear.

#110 Davers

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 9,886 posts
  • Facebook: Link
  • LocationCanada

Posted 09 October 2014 - 09:17 AM

View PostTolkien, on 09 October 2014 - 07:49 AM, said:

Even putting Elo aside, here are problems with the current system:

Is a locust equal to a firestarter? not even close. But the matchmaker calls them both 'lights' and will give the same pilot in each the same value. Regardless of who the pilot is, they will do better in the latter than the former.

Is a hunchback equal to a shadowhawk? not even close. But the matchmaker calls them both 'mediums' and will give the same pilot in each the same value. Regardless of who the pilot is, they will likely do better in the latter than the former.

Is a Quickdraw equal to a madcat? not even close. But the matchmaker calls them both 'heavies' and will give the same pilot in each the same value. Regardless of who the pilot is, they will likely do better in the latter than the former.

Is an awesome equal to an atlas? not even close. But the matchmaker calls them both 'assaults' and will give the same pilot in each the same value. Regardless of who the pilot is, they will very likely do better in the latter than the former.

Is an atlas bristling with machine guns and flamers equivalent to an atlas with real weapons? No, but the matchmaker calls them both 'assaults' and will give the same pilot in each the same value. Regardless of who the pilot is, they will do better in the latter than the former.

Is an atlas with real weapons but no modules equivalent to an atlas with real weapons, artillery, cool shot, two good mech modules, and two good level 5 weapon modules? No, but the matchmaker calls them both 'assaults' and will give the same pilot in each the same value. Regardless of who the pilot is, they will do better in the latter than the former.

Is any mech with fully unlocked skills equivalent to the identical mech with nothing unlocked? No, but the matchmaker calls them identical and will give the same pilot in each the same value. Regardless of who the pilot is, they will do better in the latter than the former.

There's a very strong argument to convert the matchmaker over to use Battlevalue and treat our individual skill level as just another BV item like it is treated in tabletop - hell keep the Elo scoring for the pilot just scale it to BV for the purposes of balancing matches.


Actually the MM tries to find mechs close to your tonnage. So it will try to match a Locust against another Locust before it then looks for Commandos, then Spiders, etc. The problem is that PGI has not implemented a system where one would want to run a Locust over any other light mech.

If you consistently run bad builds or are always using un-Elited mechs, then your Elo will drop over time. Then you won't have to face players with good builds anymore. :D

Also, Elo is NOT the Matchmaker. It is only part of the Matchmaker. There are other values PGI uses to balance matches. Many people speculate that match score is used as well. But PGI doesn't talk about it to help keep people from gaming the system.

#111 IceSerpent

    Member

  • PipPipPipPipPipPipPipPipPip
  • 3,044 posts

Posted 09 October 2014 - 09:27 AM

View Post1453 R, on 09 October 2014 - 08:50 AM, said:

Which only goes to illustrate my point. You honestly expect Piranha to accurately identify and program a correct Battle Value scale for each 0.01% of heat efficiency a 'Mech gains?


Why would you need that level of precision? On a side note, if you really need that precision for some weird reason, you can simply assign X points of BV to 0.0001 points of heat efficiency.

Quote

Assigning a BV point percentage to every last single tiny detail that can affect the outcome of a match is impossible. It. Is. Impossible. It is not a thing that is possible for sane human beings to accomplish. IT CAN'T BE DONE.


The fact that you don't know how to do something doesn't necessarily mean that task in question is impossible.

Quote

The only even remotely viable way for some sort of BV system to work is if it tracks usage. Usage, and nothing but usage.


This is one of the BV schemas on the table, was suggested by Roland.

#112 Ghogiel

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • CS 2021 Gold Champ
  • CS 2021 Gold Champ
  • 6,852 posts

Posted 09 October 2014 - 09:29 AM

View PostDeltron Zero, on 09 October 2014 - 08:20 AM, said:


They did not, they put in private matches that do not earn rewards.

A lobby system is a list of games that people made. Players can join one they find interesting. Once there are enough players and everyone is ready it can be started. Once the game is complete, it can rollover into another games. Perhaps the players can vote for a new map, or the original creator can just pick one. Players would be able to leave and new ones can join. Most importantly, this would allow you to build a consistent relationship with players over the course of a few games. If it turns out the other players are too good, or not good enough, you can drop out and join a different one.

Private lobbies are everything you asked for really, just invite only. Even if they make an open lobby system (UI component that a user can browse through the open lobbies in session and join them) we probably won't get rewards for those for the same reason we don't get rewards in private lobbies now.

#113 Tolkien

    Member

  • PipPipPipPipPipPipPipPip
  • Giant Helper
  • Giant Helper
  • 1,118 posts

Posted 09 October 2014 - 09:31 AM

View PostDavers, on 09 October 2014 - 09:17 AM, said:


Actually the MM tries to find mechs close to your tonnage. So it will try to match a Locust against another Locust before it then looks for Commandos, then Spiders, etc. The problem is that PGI has not implemented a system where one would want to run a Locust over any other light mech.

If you consistently run bad builds or are always using un-Elited mechs, then your Elo will drop over time. Then you won't have to face players with good builds anymore. :D

Also, Elo is NOT the Matchmaker. It is only part of the Matchmaker. There are other values PGI uses to balance matches. Many people speculate that match score is used as well. But PGI doesn't talk about it to help keep people from gaming the system.



[Citation Needed] - Last I read on the issue the match maker only takes into account the weight class (e.g. light, medium, heavy or assault). It doesn't take into account the actual tonnage nor whether it is a clan or IS mech, nor whether you have the skills unlocked or not, nor whether you have modules loaded or not, nor whether you have any weapons armor or ammo.

If you have a link that says otherwise, please do share it.

Edited by Tolkien, 09 October 2014 - 09:52 AM.


#114 Killstorm999999

    Member

  • PipPipPipPipPip
  • 196 posts

Posted 09 October 2014 - 09:44 AM

View PostGhogiel, on 09 October 2014 - 09:29 AM, said:

Private lobbies are everything you asked for really, just invite only. Even if they make an open lobby system (UI component that a user can browse through the open lobbies in session and join them) we probably won't get rewards for those for the same reason we don't get rewards in private lobbies now.



The invite only nature is the main problem. Its makes it more difficult to get to meet and get to know random players: you can't invite people you don't already know.

As far as rewards are concerned, if game lobbies and rewards are incompatible, then that's just one more flaw of the F2P model. I guess I wouldn't mind not having the rewards if there was no other way.

#115 Kubernetes

    Member

  • PipPipPipPipPipPipPipPipPip
  • The Blazing
  • The Blazing
  • 2,369 posts

Posted 09 October 2014 - 09:53 AM

Some of the objections don't seem very convincing. What if you drop on Alpine and the other team has 4 LRMs and lots of ECM? That says nothing about individual skill. True, but that would only be an issue if a significant percentage of your drops were like that. If you drop hundreds of times, those weirdly unbalanced matches even out. My issue is that they make up teams using average team Elo, so you get weird mixes of elite badasses and total novices. Why can't they match up teams where all pilots fall into a comparable Elo range?

#116 Duke Nedo

    Member

  • PipPipPipPipPipPipPipPipPip
  • CS 2023 Top 12 Qualifier
  • CS 2023 Top 12 Qualifier
  • 2,184 posts

Posted 09 October 2014 - 09:57 AM

View PostFire and Salt, on 09 October 2014 - 07:09 AM, said:

Random, you say? Ever heard of the law of large numbers?

Wrong. It will probably cause your team to lose a little more often than had you participated. The more trials you conduct, the more likely this will show. Again, law of large numbers.

But thanks for ignoring my detailed experiment posted above done using a new account and a trial nova. I won my first 11 games, about 20 out if the first 25, and this was in the solo queue where I was only 1/12th of the team.


He does have a point, your sample is small. It would be interesting if you would repeat the same experiment three times or more, and post screenies of every match-result. Then analyze your contribution in every game, your team contributions and the match composition. Did you carry every game, or did your artificially low elo cause you to also be carried by one or more additional competent players? I can easily see the latter causing stomps if it indeed happens. Would be interesting to see...

Edited by Duke Nedo, 09 October 2014 - 09:58 AM.


#117 Trip Hammer

    Member

  • PipPipPipPipPip
  • 135 posts

Posted 09 October 2014 - 10:06 AM

I don't really think that ELO as it stands does a very good job as it doesn't really encompass the actual situation for each player and Mech.

I would like to see a system that uses a BV system that is taylored to this game specifically combined with data taken from the players stats with that mech.

IMO (and thats as the old saying goes, opinions and ******** everybodies got one) BV should be calculated as follows (Armor total+Agility Value+(Ratio of Firepower to Heat Efficiency))

I'm not sure if there should be an extra variable for Average Engagement range or mech speed or not but as this is just a pipedream.....

They could then look at the average damage, kills, Kill Assists, Deaths and THEN Win/Loss values (weighted in that order) for the Player Skill level.

I think that this method would make for a better overall value of a player. I don't know how it would affect newer players however. and I dont know if it would stop a team from getting a total newb on their team.

Lust my two cents worth.

#118 Shibas

    Member

  • PipPipPipPipPipPip
  • 250 posts

Posted 09 October 2014 - 10:38 AM

Well, ELO itself would work if there were consistent teams or smaller groups.  Trueskill would probably work much better as a player evaluation for this game.
BV is a terrible choice for balance.  Yes, it would sort of balance mech designs in that they have an arbitrary number attached to them but no indication of viability.  Unfortunately it can be skewed because of weapons loadout.  Take for example a Raven -4X and a Raven -3L. The -3L being the obvious dominate chassis;  If you put an AC/20 on the 4x, you actually come out roughly close or higher than the BV of a Raven -3L with an LPL +2xML and ECM.  One of these is a major threat, the other is a gimmick that helps to get through XPing other variants.  That's not really a balanced setup.

Edited by Shibas, 09 October 2014 - 10:39 AM.


#119 Artgathan

    Member

  • PipPipPipPipPipPipPipPip
  • Knight Errant
  • Knight Errant
  • 1,764 posts

Posted 09 October 2014 - 10:52 AM

First Step to Understanding Elo: Understanding the meaning of expected value.

Hint: It has something to do with not focusing on specific instances of something, and instead focusing on the "bigger picture". Yes, an individual match might be crummy. However over the long run, it will balance out.

#120 Tolkien

    Member

  • PipPipPipPipPipPipPipPip
  • Giant Helper
  • Giant Helper
  • 1,118 posts

Posted 09 October 2014 - 11:33 AM

View PostArtgathan, on 09 October 2014 - 10:52 AM, said:

First Step to Understanding Elo: Understanding the meaning of expected value.

Hint: It has something to do with not focusing on specific instances of something, and instead focusing on the "bigger picture". Yes, an individual match might be crummy. However over the long run, it will balance out.


Counterpoint - when someone switches between a great mech and a crummy mech, e.g. leveling locusts or something else terrible in their weight class, their Elo will slew down until it converges on *you are driving a trash mech* then it will go back up slowly when they stop using bad stuff.

The problem here is that the skill of the player hasn't changed at all, yet Elo is giving very wrong predictions of your odds for a long time after you switch, and then for a long time after you switch back.

This is because it is ignoring so much information (chassis, loadout, modules, skills) that the presumption the E<> (Expected value of the function) is destroyed. It has a substantial bias introduced by equipment choice.

Note: Before you dismiss the effect of 1 player on a team going from a good mech to a terrible mech read up on the lanchester square principle.

Edited by Tolkien, 09 October 2014 - 12:19 PM.






1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users