Taurich, on 04 February 2013 - 08:14 AM, said:
sqrt seemed like the cleanest solution, I tried a few other formulas with constants but they were very very ugly
Throw some more formulas out so we can apply them to the stats given in this thread and we'll see what gives the cleanest and most useful results
edit: Ideally you want something that will prevent a 'Player E' situation. That player has far too high a score for his stats imo. He certainly should not be rated as the highest quality player of the group
Or his team lost the cap war a lot. Actually Player E as long as he is not KDR manipulating, is supposed to have a high score. The whole point of the stat is to value the KDR as a primary statistic. Because if those stats are accurate then Player E maintained a 6 KDR over 20 losses or 1.5 kills per match with only 1 death every 4th match. Which means either he's really good or his opponents really sucked.
Ok, well. from a theoretical standpoint.
-There should exist a standard that balances players with vastly disparate overall time played.
-This standard should not reward players who play for much longer than the average player per day
-and should not punish the casual player who may only be able to play for a few hours a week.
-This standard should also not reward players who sacrifice their team for their own solo goals,
-nor punish skilled players for the failures of their team.
-On the other hand it should accurately reflect individual player skill even under the worst conditions.
The only variables available are total wins, total losses, total kills, and total deaths. KDR is merely a pre calculated ratio of kills over deaths.
For the first/second/third conditions both wins over losses and kills over deaths provide ratios that balance the available time a player has compared to all the players.
The 4th and 5th conditions are broken by just using either the KDR or the WLR alone. However.. most other ratios don't really make sense. What would kills over wins mean(kills per win)? or even kills over wins plus losses (Kills per match), Losses over deaths? I suppose if you wanted to know how likely a person was to sacrifice for the win.
Kills over losses though.. that makes sense. Someone who has a lot of kills but loses a lot is generally leaving his team out to dry. Someone with very few kills but very few losses may either be being carried a lot or is a really good team player filling an important roll that ensures success. Having a lot of kills and few losses would be the ideal condition but it doesn't reflect the durability and survivability of the player. And of course a low number of kills and a lot of losses is self explanatory.
So, Kills over losses.. good ratio.
Using total wins as a comparison doesn't work for me. Keepign track of wins, eh, really there is no point. It doesn't show anything. Someone could have a huge number of wins and a huge number of kills and few deaths.. All that tells me is the player hasn't been challenged. Tracking wins as a statistic doesn't accurately reflect how the player may do in the worst of conditions.
The KDR. I like it.. it's ok. Doesn't say much except as a reflection of the players survivability.
So.. you take the KLR times the KDR.. there you go the HC Stat. That to me is a meaningful comparison. (OBTW thats Total Kills times the KDR over the total losses)
The problem with just using the KDR over losses is there is no balance between casual and keyboard warrior players. Using the KLR with the KDR balances the two extremes of time played.
So, to use your conditions. but check same conditions under higher multiples.
Player A: 20 Kills, 5 Death, 10 Wins, 10 losses:
Player A10: (200 kills, 50 deaths, 100 wins, 100 losses)
A: KLR 2. KDR 4. KDR over Losses .4. HC Stat 8
A10: KLR 2. KDR 4. KDR over Losses .04. HC Stat 8
So the differences between the two player of equal skill but unequal play time can't be reflected in KDR over losses alone. There has to be a balancing factor. By using the KLR times the KDR it balances the play time. However.. this does mean an exponential expansion of the statistic instead of a linear expansion. Or at least a greater than linear expansion.
Our current discussion is talking about a way to decrease the effect of the greater than linear expansion/exponential expansion caused by using the total kills times the total kills in this statistic by taking the root of the kills over deaths.
Which is taking the KLR times the square root of the KDR, instead of just using the KDR. This doesn't make sense because the KDR is a linear ratio of equal value to the KLR. There is no point to taking the root of it. That only exponentially reduces the effectiveness of the KDR in the statistic in comparison to the KLR.
The KDR has full and equal value when compared to the KLR. Taking the root of it means that it has a drasticly reduced value compared to the KLR.





















