test content
Logo
What is the Arc Client?
Install Arc
Elo system is fundamentally flawed, fix inside
zaodonn
Member Posts: 109
I will demonstrate how the Elo system doesn't work for NWO.
Since we don't know the numeric range of Elo, I will use "1-1000" for purposes of demonstration only.
Say you have 4 "premade" teams as follows:
Teams 1+2: 8000 gear score, all blue gear, never PvPed before.
Teams 3+4: 15k+ gear score, all T2.5 (tenacity) gear, all Rank 10s/Perfect enchants, lots of PvP experience.
If Teams 1 + 2 are matched with each other and, lets say, are fairly evenly skilled, then it is reasonable to say they will have about a 50/50 win/loss ratio when playing each other. So, for purposes of my example, they are ranked 500 each.
Same is true for Teams 3 + 4 if they play each other. Even skills, even ratings: also rated 500.
So, here is the flaw.
Since all 4 teams have a 500 rating, the game will now happily match 1 vs 3, 2 vs 4, whereupon Teams 1 + 2 will get eaten for breakfast.
We originally complained to you, Cryptic/PWE, about unfair matches. You decided to implement Elo. As you can now see, Elo does absolutely NOTHING to prevent unfair matches, because Elo only ranks based on skill alone, and does NOT account for gear score.
Possible fix:
- each time a player engages in combat during a match record their gearscore at that moment
- at the end of the match, take an average of these gearscores which you recorded throughout the match
- take that average gearscore value and use it as an extra factor when calculating Elo. Lower gear scores would lower the Elo, higher GS would raise it.
In my examples above, Teams 1+2's Elos would drop to, lets say 250 (half of what Elo would normally be) due to their horrible gear scores, and Team 3+4's Elos would go up to, lets say, 750 (+50% of normal Elo) due to their absolutely maxxed out GS.
This is impossible to "rig" or "scam", unless a team literally wears and KEEPS ON poor gear throughout an entire match. It also ensures that these horrible mismatches which are occurring now will cease, and our ORIGINAL COMPLAINT about uneven matches is actually addressed.
Since we don't know the numeric range of Elo, I will use "1-1000" for purposes of demonstration only.
Say you have 4 "premade" teams as follows:
Teams 1+2: 8000 gear score, all blue gear, never PvPed before.
Teams 3+4: 15k+ gear score, all T2.5 (tenacity) gear, all Rank 10s/Perfect enchants, lots of PvP experience.
If Teams 1 + 2 are matched with each other and, lets say, are fairly evenly skilled, then it is reasonable to say they will have about a 50/50 win/loss ratio when playing each other. So, for purposes of my example, they are ranked 500 each.
Same is true for Teams 3 + 4 if they play each other. Even skills, even ratings: also rated 500.
So, here is the flaw.
Since all 4 teams have a 500 rating, the game will now happily match 1 vs 3, 2 vs 4, whereupon Teams 1 + 2 will get eaten for breakfast.
We originally complained to you, Cryptic/PWE, about unfair matches. You decided to implement Elo. As you can now see, Elo does absolutely NOTHING to prevent unfair matches, because Elo only ranks based on skill alone, and does NOT account for gear score.
Possible fix:
- each time a player engages in combat during a match record their gearscore at that moment
- at the end of the match, take an average of these gearscores which you recorded throughout the match
- take that average gearscore value and use it as an extra factor when calculating Elo. Lower gear scores would lower the Elo, higher GS would raise it.
In my examples above, Teams 1+2's Elos would drop to, lets say 250 (half of what Elo would normally be) due to their horrible gear scores, and Team 3+4's Elos would go up to, lets say, 750 (+50% of normal Elo) due to their absolutely maxxed out GS.
This is impossible to "rig" or "scam", unless a team literally wears and KEEPS ON poor gear throughout an entire match. It also ensures that these horrible mismatches which are occurring now will cease, and our ORIGINAL COMPLAINT about uneven matches is actually addressed.
Post edited by zaodonn on
0
This discussion has been closed.
Comments
Incorrect.
Teams 1+2's "skill" isn't less just because the game matched them vs Teams 3+4. If their Elo goes down after getting annihilated by Teams 3+4, then they will play against other lower gear teams, and win, and their Elo will go back up, whereupon the game will match them vs the elite teams for more crushing.
Its FLAWED.
Problem solved within 1 more match. This is in fact exactly how the system is supposed to work.
If you consistently beat teams, you'll go up in score, if you consistently lose, you'll go down. Eventually you'll reach a point where you're up against teams who are comparable, and your score won't change significantly until you get better (or worse).
See note above on how you are also wrong.
You're also using a reductio ad absurdem approach with only 4 teams of two highly disparate skill levels, which is about the worst possible scenario to field test a matchmaking system, and yet it STILL WORKS.
Give your hypothetical system 5 matches and it'll exclusively pitch 3 vs 4 and 1 vs 2.
Absolutely false.
After 5 matches, it will. Then, after 10 MORE matches, it will start pitting 1+2 vs 3+4 again!
Look, if you don't understand Elo, please don't reply to the thread. Elo was made for Chess and other contests where there isn't any secondary factors to winning, its pure skill only. NW PvP has GearScore, a SECOND factor to winning. Its not like the White chess pieces have 15k GS and the black pieces have 8k....
Again, if this goes over your head, stop replying.
Elo is flawed because Elo alone CANNOT (by design) account for GearScore.
It doesn't need to. If your gearscore comes from useful stats that improve your performance, then you'll be more successful. If you aren't, then your ELO will drop until you start facing people that you can actually beat, whatever your gear score.
Yes. This is correct. And not an error.
The problem here is that you don't understand the metric. It's not measuring skill, OR gear, it's measuring win/loss ratios. Winning (or losing) is dependent on team, skill, and gear, all very nebulous, very variable concepts. Win/loss is not. Win/loss is a straight, easily measurable number.
If you generate a score based on win/loss, and aggregate it for each team before the match, you can then uprate the score of the winners by a little (if they beat a team with a lower aggregate score) or a lot (if they beat a team with a higher aggregate score), you eventually end up with a scenario where people are pitched against people they won't kerbstomp, and who won't kerbstomp them.
THIS WILL HAPPEN REGARDLESS OF GEAR. If a player is amazing but has crappy blues, they will end up fighting players with perhaps less skill but better gear, and THIS IS FINE. They deserve to fight people who present a challenge.
Using this system makes it essentially gear independent, and this is a very very good thing.
This is just hilariously misguided and the OP should expect to be laughed at unmercilessly.
ELO is a measure of winning/losing probability against other players. It implicitly takes into account all factors (GS included) so long as this does not vary considerably for the individual involved between matches.
First of all,
Stating someone is wrong when he does no agree with you and/or telling people not to post opposing arguments doesn't make you right.
Further more, you are the one that misunderstand how Elo works.
Elo doesn;t care about your GS, if you remmber all the possible chess moves, a big blue supercomputer whispering your ear or you hired a mobster to threaten your opponent to assure your victory. All the factors are included in your rank and only measuring your probability to win against another ranked person.
A very skilled person in blue is the best match to very geared less skilled player as long as they have 50/50 percent to win.
The only drawback for Elo rank stabilization is that it meant for single people or constant teams and not changing teams.
I think that you error comes from the assumption that the rank reduction or addition is constant, it is not, victory of higher ranked person will give you more points while a loss will reduce less points than a loss to equal rank.
Aelar Hawkwind - Archer
Karrin Feywinter - Mistress of Flame
Errin Duskwalker - Executioner
Darquess - Soulbinder
Under 10 one group, over 10 another and over 15 the third. Seems like that wouldve worked ok. I know you can gear swap, but oh well.. I cant believe people would go in naked to face under 10s so they can just womp them if they were already 15+ normally.
ELO can compensate for these issues a little bit, because those variables ultimately will affect the calculated skill level. The rating is absolutely useless to determine the best PVP player though because it's a single-player rating in a group context. Two players with similar gear and similar skill will end up with vastly different ELOs if one will be exclusively pugging and the other be part of premades.
Secondly, it's not fair to everyone, but that's not the challenge it has to face. The question is: Is it fairer than before?
You are claiming that the system works because a lower gear score player will GET CRUSHED HORRIBLY SEVERAL TIMES and their Elo will lower until they face people with their same gear score. And then, if they start winning in that lower bracket, and their Elo goes up, they will START GETTING CRUSHED AGAIN until they are knocked back down again.
This is why its FLAWED. If you don't understand that, for the 4th time, stop replying to the thread.
Players did not ASK for Elo. They asked for a way to MATCH players based on GEAR so that the match is all about who has better SKILL. Instead, we got Elo. Elo doesn't WORK. It regularly matches players in HOPELESS GAMES until they get crushed down to a lower Elo. That is a COMPLETE FAILURE of what was asked for.
What is being described is nothing other than players not being accurately rated. The more everybody plays the better the matches will be. And I do mean everybody.
As time goes on and people get more accurately rated this issue will diminish but for now it will be a bit of a rollercoaster as players try to separate themselves from the pack. There are people rated lower and higher than they should be and that is what is happening and what is being described.
There are plenty of preexisting threads to continue this discussion though. No need for another thread to split it even more.