Jump to content

Elo Vs Rpi Player Ratings

elo is bad mkay

46 replies to this topic

#1 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 03 May 2018 - 01:01 AM

In 1996 I wrote a Ranked Percentage Index rating system for Mechwarrior because ELO is terrible. Let me explain why!
ELO only takes into account wins-draw-lose. It assumes performance is a random distribution for each player and a change in ELO is based on the previous rankings of the players based on this limited set of data.
The RPI system takes into account TONNAGE/Mech Used, player ranking, opponents ranking, opponent's opponent rankings and damage done!
So, let's say an Anni 1x fights an Anni 2x..and the players are the same person with the same ranking...the mech chassis that is performing better across the game is factored into the match score. So, the 1x loses but may still gain RPI if the 2x is the meta mech if it did a good amount of damage compared to the normal match outcome.

#2 Monkey Lover

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 7,918 posts
  • LocationWazan

Posted 03 May 2018 - 01:28 AM

I agree it could be improved but when the wait times we have now no balance system can be used anyway. Most the time its only 2 people in the q.

#3 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 03 May 2018 - 05:12 AM

http://www.mechreg.org/Solaris.htm#RPI

Here is more information from The old website. Remember it was the 90s and kills were used instead of damage because each battle had rounds.

While this would affect matchmaker, the reason is accurate player rankings

#4 Throe

    Member

  • PipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 1,027 posts

Posted 03 May 2018 - 08:43 AM

[redacted]

Edited by Throe, 12 December 2018 - 05:57 PM.


#5 Monkey Lover

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 7,918 posts
  • LocationWazan

Posted 03 May 2018 - 01:24 PM

View PostThroe, on 03 May 2018 - 08:43 AM, said:

Individual 1v1 fights between highly mismatched opponents should have a fairly clear outcome, will probably only decrease the forecasted loser's SSR by low single digits, and the match should be over fairly quickly anyway. On the other hand, if the forecasted loser wins instead, they stand to gain a comparatively large amount of SSR.

So although it may make some players angry that this type of match up occurs at all, the system still works.

Not trying to say anything about Surn's comments about RPI here. I think he's got something here, and he clearly understands the subject better than I do.


I agree with what you're saying but its not helping the people from getting smashed. I just had a match with a elo in the 1300. Poor guy was at 6 in 33. That div i think im about flip this.

At what point is he going to quit? Have other lower skilled people already quit? I wonder if the battle value system would have gave him a shot.

Edited by Monkey Lover, 03 May 2018 - 01:25 PM.


#6 Throe

    Member

  • PipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 1,027 posts

Posted 03 May 2018 - 01:40 PM

[redacted]

Edited by Throe, 12 December 2018 - 05:58 PM.


#7 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 12 May 2018 - 06:48 PM

There is another bad side to the ELO coin...

once you get in the top 25 or whatever, it discourages you from risking your score. RPI is fair, so fair that a great player could use a really bad mech or find another way to equal the playing field and lose without risking his ranking.

For example, in div 1 I lost 100total ELO using a new mech and working out the config. I eventually got it right and had to fight back my ELO which took several hours and a couple angry looks from my wife.

RPI avoids all that by using available data to realize I was crap in that mech and treat me as such.

Edited by Surn, 12 May 2018 - 06:49 PM.


#8 Monkey Lover

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 7,918 posts
  • LocationWazan

Posted 13 May 2018 - 02:08 AM

View PostSurn, on 12 May 2018 - 06:48 PM, said:

There is another bad side to the ELO coin...

once you get in the top 25 or whatever, it discourages you from risking your score. RPI is fair, so fair that a great player could use a really bad mech or find another way to equal the playing field and lose without risking his ranking.

For example, in div 1 I lost 100total ELO using a new mech and working out the config. I eventually got it right and had to fight back my ELO which took several hours and a couple angry looks from my wife.

RPI avoids all that by using available data to realize I was crap in that mech and treat me as such.


haha i had some big elo drops from disconnecting. Really I dont want a higher elo. Im already getting skipped every time there is 3 players in the Q.

#9 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 13 May 2018 - 07:01 PM

View PostMonkey Lover, on 13 May 2018 - 02:08 AM, said:


haha i had some big elo drops from disconnecting. Really I dont want a higher elo. Im already getting skipped every time there is 3 players in the Q.


3 players someone gets skipped, but with a more accurate system, it would make a better decision.

#10 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 13 May 2018 - 07:05 PM

With a developed RPI system, each mech with a mechbay would have a strength of schedule after a min number of matches, my league used 6.

It can take into account all types of matches or have a unique sos for each game mode.

disparities like disconnects and afk can be factored into results, and even matches with unequal teams can be fairly evaluated.

Edited by Surn, 13 May 2018 - 07:08 PM.


#11 BTGbullseye

    Member

  • PipPipPipPipPipPipPipPip
  • The Solitary
  • The Solitary
  • 1,540 posts
  • LocationI'm still pissed about ATMs having a minimum range.

Posted 13 May 2018 - 09:13 PM

Honestly, this should be used for more than just Solaris. (as long as it isn't heavily based on win/loss ratio, which outside 1v1 is very inaccurate for individual skill level assessment)

#12 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 13 May 2018 - 11:40 PM

View PostBTGbullseye, on 13 May 2018 - 09:13 PM, said:

Honestly, this should be used for more than just Solaris. (as long as it isn't heavily based on win/loss ratio, which outside 1v1 is very inaccurate for individual skill level assessment)


Agreed, win/loss is hard to factor in, but let's consider what factors we might use...

win/loss
tons used
average tons opposing
damage done
average damage opposing
average damage done
Chassis and Model used (rating based on Strength of Schedule)
Average Chassis and Model rating opposing
Kills or Kills most damage
components destroyed
Deaths or Opponent average kills most damage
components lost
Number of Mechs Opposing
Number of Mechs Used
Pilot SOS or RPI
Average Opponent SOS or RPI

what might I be missing?

Anyhow, these factors are combined to create ratios then these ratios are weighted.

Eventually, you get Strength of Schedule *Win\Loss = RPI

However, I have a trick to make the database less intensive that enabled me to record 8v8 matches with many fewer calculations and transactions on the database with 90's era systems.

Edited by Surn, 13 May 2018 - 11:40 PM.


#13 Liveish

    Member

  • PipPipPipPipPipPipPip
  • CS 2022 Referee
  • CS 2022 Referee
  • 838 posts
  • LocationDarwin

Posted 14 May 2018 - 12:05 AM

ELO = If you are good and play you will go up

The problem with S7 is not ELO, just the pool of players is too low.

#14 BTGbullseye

    Member

  • PipPipPipPipPipPipPipPip
  • The Solitary
  • The Solitary
  • 1,540 posts
  • LocationI'm still pissed about ATMs having a minimum range.

Posted 14 May 2018 - 01:29 AM

View PostSurn, on 13 May 2018 - 11:40 PM, said:

what might I be missing?

Number of mechs that damaged the player. (getting focussed down by 9 enemies at once shouldn't affect your rating as badly)
Weapons used. (was he in the meta, or using UAC20's on a MCII-B with no backups?)
Was the player with their team regularly? (it's a match score point already in the game)
AMS support effectiveness.
For Conquest, were you capping points? (low on other stuff shouldn't negatively affect your score if you're in a fast mech capping)
Skill nodes used.
Weapon accuracy. (stats for that are on the forum profile, so they're already recorded)
Rate of improvement. (like if you started overheating a lot less because you spent some time in the Training Grounds practicing, and you shouldn't be penalized because you used to be a potato if you aren't one now)
Framerate comparison. (playing at 10-25 FPS is MUCH harder than playing at 45+ when dealing with non-homing weapons)
Effective uses of consumables. (scoring hits with strikes, or spotting with UAVs)
Counter-ECM effectiveness.
NARCing/TAGging.
Comms usage. (this would be really hard to implement, and impossible for voice)
LOCKING TARGETS!!! (this really should be a big one for the QP version of this)
Firing on locked targets vs firing on a target other than what you have locked. (this would have to be an unbelievably minor affect due to the various reasons to not change your lock in combat, probly simpler to ignore it)
Scouting locks. (first lock on a target)

That's all I can think of right now... Maybe I'll think of more later.

Edited by BTGbullseye, 14 May 2018 - 01:33 AM.


#15 Vellron2005

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Blood-Eye
  • The Blood-Eye
  • 5,443 posts
  • LocationIn the mechbay, telling the techs to put extra LRM ammo on.

Posted 14 May 2018 - 02:14 AM

If you ask me, ELO works flawlessly for Solaris.. matches are equally matched and close.

For modes other than solaris, I think we need a battle value system which looks at not only player statistics, but mechs, weapons and equipment installed too..

The same player is not equally good in a light as in an assault, nor is he equally good in an SRM assault as a ERLL assault. All of these factors, each weapon, engine, ecm, JJ and computer should have a battle value associated with them, along with player's stats, and then the MM should compare the totals and match similar numbers.

#16 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 14 May 2018 - 05:56 PM

View PostBTGbullseye, on 14 May 2018 - 01:29 AM, said:

Number of mechs that damaged the player. (getting focussed down by 9 enemies at once shouldn't affect your rating as badly)
Weapons used. (was he in the meta, or using UAC20's on a MCII-B with no backups?)
Was the player with their team regularly? (it's a match score point already in the game)
AMS support effectiveness.
For Conquest, were you capping points? (low on other stuff shouldn't negatively affect your score if you're in a fast mech capping)
Skill nodes used.
Weapon accuracy. (stats for that are on the forum profile, so they're already recorded)
Rate of improvement. (like if you started overheating a lot less because you spent some time in the Training Grounds practicing, and you shouldn't be penalized because you used to be a potato if you aren't one now)
Framerate comparison. (playing at 10-25 FPS is MUCH harder than playing at 45+ when dealing with non-homing weapons)
Effective uses of consumables. (scoring hits with strikes, or spotting with UAVs)
Counter-ECM effectiveness.
NARCing/TAGging.
Comms usage. (this would be really hard to implement, and impossible for voice)
LOCKING TARGETS!!! (this really should be a big one for the QP version of this)
Firing on locked targets vs firing on a target other than what you have locked. (this would have to be an unbelievably minor affect due to the various reasons to not change your lock in combat, probly simpler to ignore it)
Scouting locks. (first lock on a target)

That's all I can think of right now... Maybe I'll think of more later.


Basically, any stat collected or accurately calculated could be used. Further, there could be different RPI calculations for game modes, while keeping a basic combat performance RPI.

So you could have
Combat RPI
Teamplay RPI
Game mode RPI
RPI per weights class
RPI per mech chassis and model
RPI with a class of weapons

the combinations are endless

Back in mw2 to mw4, mech commander, etc days...all we had were kills, deaths, tonnage, chassis, and number of players on each team.

Edited by Surn, 14 May 2018 - 05:59 PM.


#17 BTGbullseye

    Member

  • PipPipPipPipPipPipPipPip
  • The Solitary
  • The Solitary
  • 1,540 posts
  • LocationI'm still pissed about ATMs having a minimum range.

Posted 14 May 2018 - 06:26 PM

I like the idea of multiple RPIs, and having hard and soft limits for each for matchmaking...

Try and match as close as possible the overall combat RPI, but can be much more lenient for the other RPIs.
Teamplay RPI is far more important for FP than QP, but is useful for both. It's useless for Solaris 1v1 however.
Weapons, Classes, modes, and mechs RPIs would just be a subclass of the Combat RPI. Good for skill comparisons, but not really for matchmaking.

Combat RPI = CR
Teamplay RPI = TR
Other RPIs collectively - OR

Matchmaker rating for QP single queue = 80% based off CR, 15% based off TR, 5% based off OR.
Matchmaker rating for FP and QP group queue = 60% based off CR, 35% based off TR, 5% based off OR.
Matchmaker for Solaris = 80% based off CR, 20% based off OR.

Edited by BTGbullseye, 14 May 2018 - 06:27 PM.


#18 Naqser

    Member

  • PipPipPip
  • The Deadly
  • The Deadly
  • 60 posts

Posted 15 May 2018 - 06:16 AM

The more values you choose to implement into a skill calculation equation, the higher risk you run of easier manipulation / explotation.
The second issue is balancing the values to each other in the equation to reduce the risk of manipulation / explotation.

I acknowledge the issue we have with ELO and the many many mechs we have, and the Metas.
However if you want to win, you need to acknowledge the fact that you need to either accept the imbalance and keep doing your thing, or take whatever is doing best and do that.

Rather than introducing overly complex equations and whatnot ( which would take considerable time to balance up out in the wild to get a desirable result ), ingame mechanics could be introduced to even out the average playing field ( because that's what a skill system basically is, a number based on your outcome from a number of matches).

There could be Mech ban rotations to force players into a variety of mechs.
We could have drop decks you need to fill out. Using a mech will lock it untill all mechs are locked. So you'd need to play through all mechs each at a time.

#19 Surn

    Member

  • PipPipPipPipPipPipPipPip
  • The God of Death
  • The God of Death
  • 1,073 posts
  • Twitter: Link
  • LocationSan Diego

Posted 15 May 2018 - 02:12 PM

View PostNaqser, on 15 May 2018 - 06:16 AM, said:

The more values you choose to implement into a skill calculation equation, the higher risk you run of easier manipulation / explotation.
The second issue is balancing the values to each other in the equation to reduce the risk of manipulation / explotation.

I acknowledge the issue we have with ELO and the many many mechs we have, and the Metas.
However if you want to win, you need to acknowledge the fact that you need to either accept the imbalance and keep doing your thing, or take whatever is doing best and do that.

Rather than introducing overly complex equations and whatnot ( which would take considerable time to balance up out in the wild to get a desirable result ), ingame mechanics could be introduced to even out the average playing field ( because that's what a skill system basically is, a number based on your outcome from a number of matches).

There could be Mech ban rotations to force players into a variety of mechs.
We could have drop decks you need to fill out. Using a mech will lock it untill all mechs are locked. So you'd need to play through all mechs each at a time.


Again, already have 10years of experience with this in a MechWarrior based league. Making a combat RPI is just a matter of integrating damage done and changing from tonnage to tonnage with a chassis and model modifier.

Edited by Surn, 15 May 2018 - 02:16 PM.


#20 MischiefSC

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 16,697 posts

Posted 16 May 2018 - 08:59 AM

Except a matchmaker is matching people based on their ability to win matches - not do damage. LRMs do damage, lots of it. HGauss do less total but kill better. MGs are great for kill stealing but don't indicate how likely you are to.... You kniw.

Win the match.

There is no value, purpose or reason to build a matchmaker whos purpose is to balance odds of winning around anything but actual winning.

That's why Elo is a thing.

Edited by MischiefSC, 16 May 2018 - 09:00 AM.






1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users