184 replies to this topic

[BL00D RAVEN]

0.19
i'm a light pilot

edit: i bet i have the lowest posted score

Edited by BL00D RAVEN, 04 February 2013 - 06:10 AM.

[ConnorSinclair]

lol harlequin number, LMAO

I PUT MY NAME IN IT!

[Triggerhippy]

This metric shows nothing - it would mean more as a per weight class stat. But as xifferant classes require differant play styles someone like myself who runs all types (23 mechs in my mech bay) my med mechs performance drags down my heavies performance etc. My scouts also destroy this metric as I have a very high win ratio but rarely kill anything yet heavies and assault I can regularly bring hone 3 or 4 kills.

[Taurick]

HC Harlequin, on 04 February 2013 - 04:55 AM, said:

Well, one characteristic of this statistic is that generally people with a certain amount of skill will get more than 1 kill per match and if they do so they will generally not lose which causes the statistic to increase even more. These two compositions of individual skill are imbedded in this statistic which will cause quality players to have quite a large advantage in this stat.

No doubt, however the wide and wild range of numbers makes it somewhat counter intuitive to read. With numbers ranging from 0.4 to 40+ it's difficult to make sense of it

I'd normalise the ranges somewhat by changing the formula to

(Kills * sqrt(KDR)) / Losses

Just brings those anomalous high and low scores back towards the center a bit, and places a litrtle more weight into the match loss metric



It's not any better or worse than the current formula, it just seems to hurt my eyes less when looking at results haha

Edited by Taurich, 04 February 2013 - 06:33 AM.

[Redshift2k5]

13

[HC Harlequin]

Taurich, on 04 February 2013 - 06:30 AM, said:

No doubt, however the wide and wild range of numbers makes it somewhat counter intuitive to read. With numbers ranging from 0.4 to 40+ it's difficult to make sense of it

I'd normalise the ranges somewhat by changing the formula to

(Kills * sqrt(KDR)) / Losses

Just brings those anomalous high and low scores back towards the center a bit, and places a litrtle more weight into the match loss metric



It's not any better or worse than the current formula, it just seems to hurt my eyes less when looking at results haha

That has the same result as having the dispersion in the poll. It's eliminating the exponential, however it doesn't incorporate the total matches aspect.

[Bad Karma 308]

(1,404 * 2.28) / 597 = 5.36


But I think I agree with some of the other guys that breaking out your bell curve by weight class might prove more accurate. Some classes just bring more to the table and skill in that role would be a far better indicator. I think your bell curve might become far more apparent with that type of data.

But as people jump around in mechs and MWo doesn't offer up those type of stats, I couldn't imagine how to accurately acquire that data set.

The other thing that occurred to me is that even now, how can you ensure that numbers people have already posted are accurate either? A good lesson I've learned is to never totally trust people, even less those from the internet.

Edited by Bad Karma 308, 26 February 2013 - 11:01 PM.

[De La Fresniere]

Doesn't make any sense to me. Why would you *divide* by losses?

Maintaining a good K/D ratio is harder if you play with unskilled teammates, so having a good K/D ratio and a high loss rate should yield a better score than having the same K/D ratio and a high win rate.

[Taurick]

HC Harlequin, on 04 February 2013 - 06:37 AM, said:

That has the same result as having the dispersion in the poll. It's eliminating the exponential, however it doesn't incorporate the total matches aspect.

?

It is identical to your current formula except less weight is given to KDR
It measures the same data sets with the same references

The only difference is it is less prone to outliers, ie you will see a mean and median develop earlier, and anomalous scores will be easier to see

Edited by Taurich, 04 February 2013 - 06:47 AM.

[HC Harlequin]

De La Fresniere, on 04 February 2013 - 06:38 AM, said:

Doesn't make any sense to me. Why would you *divide* by losses?

Maintaining a good K/D ratio is harder if you play with unskilled teammates, so having a good K/D ratio and a high loss rate should yield a better score than having the same K/D ratio and a high win rate.

How would losing more than winning be better than winning more than losing?

Taurich, on 04 February 2013 - 06:45 AM, said:

?

It is identical to your current formula except less weight is given to KDR
It measures the same data sets with the same references

The only difference is it is less prone to outliers, ie you will see a mean and median develop earlier, and anomalous scores will be easier to see

no. actually it's not, because you are incorporating the deaths into the root.

[Alistair Winter]

@De La Fresniere: I would rather reward players who find a way to win instead of just trying to improve their own KDR. IMO, the ultimate test is whether or not you win battles. I'm more impressed by Puggers with a positive WLR (Win Lose Ratio) than a positive KDR.

Unfortunately, as others have pointed out, this statistic isn't very useful at all. First of all, you can't demonstrate a bell curve if the scale is off. The Bell curve could conceivably be seen peaking at 0.5, or even 0.1.

On top of that, there's the issue of whether the people voting represent the average. I would say it's 99% probably that the people voting are much, much better than the average player. The average player probably doesn't post a whole lot on the forum, for a start.

It's a very interesting question, but I don't think you'll find your answer this way.

[Taurick]

HC Harlequin, on 04 February 2013 - 06:48 AM, said:

no. actually it's not, because you are incorporating the deaths into the root.

I haven't introduced any additional data sets, I have simply reduced the weighting of a data set you had already incorporated

[HC Harlequin]

Taurich, on 04 February 2013 - 06:51 AM, said:

I haven't introduced any additional data sets, I have simply reduced the weighting of a data set you had already incorporated

the KDR is kills over deaths. therefore the square root of the KDR is the square root of the kills over deaths, therefore you are incorporating the deaths into the root.

Just realized I mispoke. When I said losses I meant deaths. my apologies

Edited by HC Harlequin, 04 February 2013 - 06:59 AM.

[De La Fresniere]

HC Harlequin, on 04 February 2013 - 06:48 AM, said:

How would losing more than winning be better than winning more than losing?

Individual players have very little control over the outcome of the match. I could kill three times my tonnage in opponents every match and still lose 3/4 of the matches I'm in.

A player who has a good K/D ratio even if he loses every single match he's in (and thus almost always dies)? That dude is a *much* better player than a guy who happens to score a couple kills per match but almost never dies because he's usually in overwhelming wins.

If losses affect the score, they should do so positively, not negatively.

[Bad Karma 308]

HC Harlequin, on 04 February 2013 - 06:58 AM, said:

Just realized I mispoke. When I said losses I meant deaths. my apologies

Wait one...Is that changing your original formula:
from (Kills * KDR) / losses ?
To (Kills * KDR) /Deaths?

Cause that changes what data you've already accrued.

[Broceratops]

4700 kills, 4.4 kdr 810 losses = 25.5
with deaths, i have 1054 so I guess it would be something like 21.

I feel at the current stage any statistical measure will be heavily biased though because all the positive stats go up when you do a premade, and go down when you pug. So any statistical analysis just gives you a number in between that is dependant on how much you pug and less on anything else.

Edited by Broceratops, 04 February 2013 - 07:14 AM.

[HC Harlequin]

Bad Karma 308, on 04 February 2013 - 07:08 AM, said:

Wait one...Is that changing your original formula:
from (Kills * KDR) / losses ?
To (Kills * KDR) /Deaths?

Cause that changes what data you've already accrued.

no. That was a reply to a previous post where someone suggested taking the exponential out by taking the square root of the KDR and multiplying by the total kills. I stated that would be incorporating the deaths into the root which would skew the reduction of the exponential.

Edited by HC Harlequin, 04 February 2013 - 07:37 AM.

[Mechteric]

1235 * 2.9 / 337 = 10.63


Could there be a flaw in this number in that it doesn't track your total wins or win/loss ratio?

[HC Harlequin]

De La Fresniere, on 04 February 2013 - 07:08 AM, said:

Individual players have very little control over the outcome of the match. I could kill three times my tonnage in opponents every match and still lose 3/4 of the matches I'm in.

A player who has a good K/D ratio even if he loses every single match he's in (and thus almost always dies)? That dude is a *much* better player than a guy who happens to score a couple kills per match but almost never dies because he's usually in overwhelming wins.

If losses affect the score, they should do so positively, not negatively.

However. the reason why this statistic reflects a working relationship between losses and kills is because both are relative to the total number of matches and losses puts a limit on the overall statistic. In general, even the best of players can only get 7 kills per match and still lose. However, you can get an unlimited amount of kills and wins. The KDR on the other hand is not relative to the total number of matches played. You can have the same ratio with either 100 or 10000 matches. As well the kills over losses is a ratio that allows a comparison regardless of total number matches played. KDR on the other hand is an indication of primarily player survivability, which may be an indicator of player skill. The KDR needs a second comparison to bring relevance to it. Wins, in general, are relatively meaningless as an indicator. Any player can play and win. A much better indication of player skill in adversity when you only have 4 stats to work with is obviously overall losses. However just using losses without reducing it to a ratio is again relatively meaningless. Any player can play any number of times and lose. In this case just using a total kills over total losses is unworkable because it doesn't reduce the total number of matches to a pertinant ratio (just like the KDR). I think one of the important characteristics is to have a simple self limiting statistic that is usable for comparison across the board from casual players to keyboard warriors.

[Taurick]

HC Harlequin, on 04 February 2013 - 06:58 AM, said:

the KDR is kills over deaths. therefore the square root of the KDR is the square root of the kills over deaths, therefore you are incorporating the deaths into the root.

Just realized I mispoke. When I said losses I meant deaths. my apologies

Of course I am?
How else would you reduce the weighting of KDR without introducing some messy arbitrary constant?

I don't really see your point. KDR is still taken into account, it simply has less influence over the final score

Examples will better illustrate the elegance and diversity of the sqrt function.
Player A & B show how the original formula lacks the weighting to differentiate between a valued team player and someone who simply maintins KDR
Players C & D demonstrate that KDR still has enough weight to drastically influence the final score, even in low-loss situations. Outlier reduction also shown here
Player E demostrates a fatal flaw with the unweighted original formula; that the KDR stat dominates to the extent of making it impossible to compare HC scores side by side. In this case the original formula rated Player E the best player of the list , with 20 losses from 20 matches

Player A: 20 Kills, 5 Death, 10 Wins, 10 losses:

Org Formula: (20*4)/10 = 8

Rvs Formula (20*sqrt(4))/10 = 4

Player B: 20 kills, 10 deaths, 15 wins, 5 losses

Org Formula: (20*2)/5 = 8

Rvs Formula (20*sqrt(2))/5 = 5.656

Player C: 10 kills, 20 deaths, 10 wins, 10 losses

Org Formula: (10*0.5)/10 = 0.5

Rvs Formula (10*sqrt(0.5))/10 = 0.7

Player D: 3 kills, 20 deaths, 18 wins, 2 losses

Org Formula: (3*0.15)/2 = 0.225

Rvs Formula: (3*sqrt(0.15))/2 = 0.581

Player E: 30 kills, 5 deaths. 0 wins, 20 losses

Org Formula: (30*6)/20 = 9

Rvs Formula: (30*sqrt(6))/20 = 3.67

Edited by Taurich, 04 February 2013 - 07:50 AM.