Jump to content

Match Score Breakdown (And Help Wanted) For Science!

General

88 replies to this topic

#21 XxXAbsolutZeroXxX

    Member

  • PipPipPipPipPipPipPipPipPip
  • The Stryker
  • The Stryker
  • 2,056 posts

Posted 24 August 2015 - 01:34 PM

View PostMcgral18, on 23 August 2015 - 11:59 PM, said:

A weird value...damage seemed to be 48% towards Match Score...I don't feel too confident in this one because it's just so random, but it kinda adds up (before throwing any other bonuses in it). Previously, it was 10%.


Wouldn't mind if damage were reverted back to 10% of match score.

#22 Mcgral18

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • CS 2019 Top 8 Qualifier
  • CS 2019 Top 8 Qualifier
  • 17,987 posts
  • LocationSnow

Posted 24 August 2015 - 01:51 PM

It seems a loss actually costs you 2 points, and Match Score is a full 50% of damage (which makes a hell of a lot more sense than 48%).

#23 Revis Volek

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 7,247 posts
  • Google+: Link
  • Facebook: Link
  • LocationBack in the Pilots chair

Posted 24 August 2015 - 02:12 PM

View PostMcgral18, on 24 August 2015 - 01:51 PM, said:

It seems a loss actually costs you 2 points, and Match Score is a full 50% of damage (which makes a hell of a lot more sense than 48%).



This DOES make more sense...and i dont know how me missed it.

Oh wait...yea i do. It was 3am...lolz. :P

#24 XxXAbsolutZeroXxX

    Member

  • PipPipPipPipPipPipPipPipPip
  • The Stryker
  • The Stryker
  • 2,056 posts

Posted 24 August 2015 - 02:38 PM

View PostMcgral18, on 24 August 2015 - 01:51 PM, said:

It seems a loss actually costs you 2 points, and Match Score is a full 50% of damage (which makes a hell of a lot more sense than 48%).


Neither 48% nor 50% makes sense.

In a person has a near perfect game where every shot they fire hits the center torso of an enemy. The typical score they would have is something like 4-5 kills and 400-500 damage.

To consistently achieve a high damage score in the 800-1000+ range usually means someone has bad aim. One shot hits a leg, another shot hits someones arm, another hits the side torso. They fail to consistenly hit what they're aiming for. Or they farm to some degree.

I'm not a fan of farming or deliberately manufacturing high scores. Damage being a smaller proportion of match score makes sense to me.

#25 MrJeffers

    Member

  • PipPipPipPipPipPipPip
  • 796 posts
  • LocationThe Rock

Posted 24 August 2015 - 02:50 PM

View PostMcgral18, on 24 August 2015 - 01:51 PM, said:

It seems a loss actually costs you 2 points, and Match Score is a full 50% of damage (which makes a hell of a lot more sense than 48%).

Is it the loss, or the death? I think it's the death that costs 2 (or more) points. Need a losing match screenshot set with similar damage where there is one unit on the losing team alive. Looking at the first screen shot, and assuming that the win is worth 20, it looks like being dead is -5, and being alive is +5.

Edited by MrJeffers, 24 August 2015 - 03:00 PM.


#26 Mcgral18

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • CS 2019 Top 8 Qualifier
  • CS 2019 Top 8 Qualifier
  • 17,987 posts
  • LocationSnow

Posted 24 August 2015 - 03:45 PM

View PostMrJeffers, on 24 August 2015 - 02:50 PM, said:

Is it the loss, or the death? I think it's the death that costs 2 (or more) points. Need a losing match screenshot set with similar damage where there is one unit on the losing team alive. Looking at the first screen shot, and assuming that the win is worth 20, it looks like being dead is -5, and being alive is +5.

View PostMcgral18, on 24 August 2015 - 01:51 PM, said:

It seems a loss actually costs you 2 points, and Match Score is a full 50% of damage (which makes a hell of a lot more sense than 48%).


It is the death that costs 2 points (and suicide costing 3 for a total of -5).

A Loss and Tie do not cost anything.



SCIENCE!


Also some errors with Kill Most Damage Dealt, in that it doesn't give you any points, and actually seems to cancel out Hit and Run (1 point) bonuses.



Nearly all the preliminary data is wrong...Assists are worth 3.

#27 MrJeffers

    Member

  • PipPipPipPipPipPipPip
  • 796 posts
  • LocationThe Rock

Posted 24 August 2015 - 04:08 PM

View PostMcgral18, on 24 August 2015 - 03:45 PM, said:


It is the death that costs 2 points (and suicide costing 3 for a total of -5).

A Loss and Tie do not cost anything.



SCIENCE!


Also some errors with Kill Most Damage Dealt, in that it doesn't give you any points, and actually seems to cancel out Hit and Run (1 point) bonuses.



Nearly all the preliminary data is wrong...Assists are worth 3.


So with that and using some of the match data I was recording this weekend I come up with values for the following things:
Kill Assist =4
Damage Done = .5
Team Damage =0
Scouting =2
Lance in formation =2
Flanking =2
Protected Medium =2
Proximity Protection =2
Win =20
Death = -2.

Edited by MrJeffers, 24 August 2015 - 04:08 PM.


#28 Lily from animove

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • The Devoted
  • The Devoted
  • 13,891 posts
  • LocationOn a dropship to Terra

Posted 24 August 2015 - 05:03 PM

much science done, now tired, time to go to bed. rest needs to be tested later.

#29 odiemoncrew

    Member

  • PipPipPip
  • Bad Company
  • Bad Company
  • 83 posts

Posted 24 August 2015 - 05:03 PM

Yea. if you want someone to help test this stuff, i have at least 39/40 days of premium time,
with another 90 days that i can activate. So i got more then enough time and quite a few
I.S. and Clan mechs of all weight classes just in case we run into some data for one side
and weight class you don't think adds up.

#30 Mcgral18

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • CS 2019 Top 8 Qualifier
  • CS 2019 Top 8 Qualifier
  • 17,987 posts
  • LocationSnow

Posted 24 August 2015 - 05:24 PM

View PostMrJeffers, on 24 August 2015 - 04:08 PM, said:


So with that and using some of the match data I was recording this weekend I come up with values for the following things:
Kill Assist =4
Damage Done = .5
Team Damage =0
Scouting =2
Lance in formation =2
Flanking =2
Protected Medium =2
Proximity Protection =2
Win =20
Death = -2.


Only difference I see is Assist (and I haven't tested all of the Protected family, but Protection Proximity was worth 2, and likely all), which we found was to be 3.

Solo Kill being worth 8, but Kill Most Damage Dealt being worth 0 (and for each KMMD, it removed a Hit and Run bonus, costing 1 Match Score).

#31 Bilbo

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Nimble
  • The Nimble
  • 7,864 posts
  • LocationSaline, Michigan

Posted 24 August 2015 - 06:06 PM

It would appear that winning the game according to the game mode plays little part in match score. 54 match score for a win on assault, no damage done, started and finished cap.

#32 Lynx7725

    Member

  • PipPipPipPipPipPipPipPip
  • 1,710 posts

Posted 24 August 2015 - 06:31 PM

Ok I bite. I'm running through the EOM I have on the new scoring, I have about 14 good complete screen caps to work from.. good enough sample size. My games are fairly consistent -- all Skirmish, 10 in LRM Hunchback and 4 in DF EBJ.

At the current moment, I'm working from the following assumptions:

1. PGI used a relatively simple weighted algo. This is because programmers generally prefer simpler setup (complex ones go boom too easily), the calculations for 24 players have to be fast at the end of game, and for QC, something that is easy to eyeball from EOM scores is desirable (but by no means necessary).

2. PGI is deliberately trying to weight the algo towards team-centric awards, particularly for lighter classes Thus, anything that requires two friendly mechs to work together to achieve will likely be weighted higher, whereas things that you can do on your own is weighted lower. However, things that can create bad behaviour in game (kills, kills assists, etc.) will not be weighted highly (because, bad behaviour).

My current set of markers are as follows, and they are not yet accurate.

Win x20
Killing Blow x1
Kill Assist x1
Solo Kill x1
Kill Most Damage Dealt x1
Damage Done x0.5
Team Damage x-5
Spotting Assist x1
Component Destroyed x1
Saviour Kill x2
Flanking x1
Counter ECM x2
Counter ECM Locked Dmg x2
Scouting x5
Brawling x1
Hit and Run x1
Lance in Formation x2
Protected Medium x2
Protection Proximity x2

Some rational -- Kills and damage-based awards have never been the focus of scoring, because they promote bad behaviour. So I don't expect PGI to have weighted them significantly, especially as damage is already a major component.

Lance/ Protected awards, as well as ECM awards, I weighted higher, because these mostly require at least 2 friendly members to work together, or to devote tonnage to team equipment (either BAP, or changing ECM mode). Saviour Kill is here because at least two members of the team must be fighting in order for this to occur.

Scouting right now is weighted very high, and is unconfirmed, because of one thing -- the entire team can only scout twelve times, and only if you don't fire on it. Thus, it's a limited award.

Team Damage though.. Looking at my data, I suspect it's a sliding scale. 1 point of damage doesn't deduct much, but if you do a lot (more than 4 points?) the deduction looks to be very high. I only got one 10 Team Damage though (UAC40 build), so not enough data.

I'm not so sure on Death penalty because it really encourages bad behaviour, but it may be in place.

So moving on.. the good thing about having sucky games is that the data is much more easier to interpret. As the simplest case, I had the following EBJ game:

Win, 1 Assist, 75 Dam; EOM 58, Calculated 58

Next two best are:

3 Assist, 481 Dam, 1 Team Damage, 2 Spots, 2 Components, 7 Lance, 13 Protected Medium, 7 Proximity; EOM 296, Calculated 296

1 Assist, 249 Dam, 1 Team Damage, 2 Components, 2 Lance, 11 Protected Medium, 4 Proximity; EOM 156, Calculated 156

And a more complex case:

1 Assist, 1 Most Damage, 400 Dam, 1 Team Damage, 4 Spots, 2 Scouts, 1 Hit and Run, 5 Protected Medium, 2 Protection Proximity; EOM 226, Calculated 226

But after which, it starts to break down.

In my sample of 14 games, aside from the above 4 games that are dead on, I have 3 with 3 points under (my calculations gave lower), and the remaining 7 games doing 10 points or more under (again, I underscored the actual).

Of the 7 "bad", I have 4 with more than 2 points of Team Damage (2, 4, 5, and 10). I also had more cases of Hit And Run, Brawling, and in two cases, actually being alive at the end of game. So I need to narrow things a bit more.

That's what I have at the moment.

EDIT: Sod. Forum software can't take tables.. reformatted.

Edited by Lynx7725, 24 August 2015 - 06:42 PM.


#33 Mcgral18

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • CS 2019 Top 8 Qualifier
  • CS 2019 Top 8 Qualifier
  • 17,987 posts
  • LocationSnow

Posted 24 August 2015 - 06:31 PM

View PostBilbo, on 24 August 2015 - 06:06 PM, said:

It would appear that winning the game according to the game mode plays little part in match score. 54 match score for a win on assault, no damage done, started and finished cap.


Each "Capturing Base" tick is worth 2 points, and the Capture Assist is worth 1.

#34 Lynx7725

    Member

  • PipPipPipPipPipPipPipPip
  • 1,710 posts

Posted 24 August 2015 - 06:45 PM

View PostMcgral18, on 24 August 2015 - 06:31 PM, said:

Each "Capturing Base" tick is worth 2 points, and the Capture Assist is worth 1.

Personally I don't like to use Conquest to do these kind of maths. Hard to get a good figure.

Assault or Skirmish is easier to get a grip on things. Skirmish is simpler still, since it's straight raw killing.

#35 Mcgral18

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • CS 2019 Top 8 Qualifier
  • CS 2019 Top 8 Qualifier
  • 17,987 posts
  • LocationSnow

Posted 24 August 2015 - 06:50 PM

View PostLynx7725, on 24 August 2015 - 06:45 PM, said:

Personally I don't like to use Conquest to do these kind of maths. Hard to get a good figure.

Assault or Skirmish is easier to get a grip on things. Skirmish is simpler still, since it's straight raw killing.


We did private matches and isolated variables...and found a weird bug with Kill Most Damage Dealt (which removes a Hit and Run bonus for each KMMD, -1 score in essence).


Most of the revised data fields seem accurate, but we've yet to do in depth testing on UAVs and Formation bonuses (just that the Proximity Protection is +2).


Scouting is without a doubt 2, and dying is without a doubt -2 (with suicide -3, totalling -5).

#36 Lynx7725

    Member

  • PipPipPipPipPipPipPipPip
  • 1,710 posts

Posted 24 August 2015 - 06:57 PM

View PostMcgral18, on 24 August 2015 - 06:50 PM, said:

We did private matches and isolated variables...and found a weird bug with Kill Most Damage Dealt (which removes a Hit and Run bonus for each KMMD, -1 score in essence).

Most of the revised data fields seem accurate, but we've yet to do in depth testing on UAVs and Formation bonuses (just that the Proximity Protection is +2).

Scouting is without a doubt 2, and dying is without a doubt -2 (with suicide -3, totalling -5).

Ok, I'd adjust later and see what other things drop out. At work. Bosses prefer I actually do work...

#37 LT. HARDCASE

    Member

  • PipPipPipPipPipPipPipPipPip
  • The Spear
  • The Spear
  • 2,706 posts
  • LocationDark Space

Posted 24 August 2015 - 07:00 PM

Saw a guy, in a loss, get 1 damage and 1 assist.

Still got 23 points.

#38 Lynx7725

    Member

  • PipPipPipPipPipPipPipPip
  • 1,710 posts

Posted 24 August 2015 - 08:29 PM

Take note I'm distrusting displayed Damage values, both Damage Done and Team Damage.

This is as we know that the game can go to fractional damage, but displays always show in integer form. We don't know whether the rounding is done before or after formula. So as long as I get close enough (within 1 match score either side) I'm "happy".

Team Damage is a problem under this scheme though.

#39 riverslq

    Member

  • PipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 443 posts

Posted 24 August 2015 - 08:32 PM

so being a scout and tagging, narcing etc is useless.
great.

#40 Lynx7725

    Member

  • PipPipPipPipPipPipPipPip
  • 1,710 posts

Posted 24 August 2015 - 08:35 PM

View Postriverslq, on 24 August 2015 - 08:32 PM, said:

so being a scout and tagging, narcing etc is useless.
great.

Not enough data at this point really. Scout looks to be good, and we don't have info on TAG and NARC, and in some sense those also count towards Counter ECM and Counter ECM Locked Damage. So there's still a point to it.





16 user(s) are reading this topic

0 members, 16 guests, 0 anonymous users