next »

[Freelander]

How can xlrstats claim to calculate skill, if it does not consider shots that missed?

Look at our slogan on xlrstats.com -> "When you make the kill, we calculate your skill"

However you're right, accuracy should be one of the factors in skill calculation but this information is generally not available in gamelogs.

In general, xlrstats should not use kills for its calculations. Almost each gametype has a different scoring system, where kills usualy do not matter at all.

Do you have an idea for this that can be used globally for all different games and gametypes?

[Courgette]

do not mix up "accuracy" and "xlrstats skill".

XLRstats skill computation is derivated from the ELO system

[grosbedo]

do not mix up "accuracy" and "xlrstats skill".

XLRstats skill computation is derivated from the ELO system

Oh, I didn't expect this plugin to use such a pragmatic method to calculate its stats.

Anyway, from Wikipedia :

Subsequent statistical tests have shown that chess performance is almost certainly not normally distributed. Weaker players have significantly greater winning chances than Elo's model predicts. Therefore, both the USCF and FIDE have switched to formulas based on the logistic distribution. However, in deference to Elo's contribution, both organizations are still commonly said to use "the Elo system".

This means that the first assumption of ELO that the performances of a player is like a normally distributed random variable is false, and so the calculation is very approximate, particularly over the long-term. Plus, this is in the context of 1vs1 games that is Chess, in our context of multiplayer gaming with several opponents (and sometimes friends) at a time, this change the calculations as well.

Anyway, the ELO model has the advantage to be simple, and theoretically less ressource consuming.

Now, the question of Gedanken still seems valid to me : the datas used for the calculation of the ELO model is of a capital importance. If only the kills are used, this miss a lot of the score datas for other gamemodes than Free for all.

Note : Im not saying that XLRstats has a bad scoring algorithm, actually I think it's one of the best, and Im not saying that a rethought of the underlying model is critical, but depending on how the model is handled, you have to know the degree of approximation that the system can produce.

[GedankenBlitz]

However you're right, accuracy should be one of the factors in skill calculation but this information is generally not available in gamelogs.

Do you have an idea for this that can be used globally for all different games and gametypes?

Given the improved log I'm (we) are working on, xlrstats could simply use the AddScore logline. This will be triggered for kills in FFA, Captures in CTF etc. So it should be a good idea to use score rather than kills, since it works for all gametypes.
Still there remains a problem with this approach: You can usually easily frag 5 enemies/minute in FFA, but you won't be able to capture the enemy's flag 5 times/minute, so these scores can not be combined into one. You'd have to record separate statistics per gametype.

Furthermore these statistics might not be comparable across servers if one of them uses Instagib or mods like that, but this is just  a sidenote.

In addition one might like to consider powerups like Quaddamage, Haste etc. since they allow easier scoring. Yet it should be valid to assume that every player has the same chance to use them.


Concerning accuracy, it is not a good idea to log every shot into a logfile. One'd have to use a different communitcation method for game<->b3 for such events.
The problem with other gametypes also applies here. While in FFA you can substitute accuracy/kills with skill, in CTF you can not. Even if a player does not shoot at all, he might be the one scoring best in his team.

[grosbedo]

Still there remains a problem with this approach: You can usually easily frag 5 enemies/minute in FFA, but you won't be able to capture the enemy's flag 5 times/minute, so these scores can not be combined into one. You'd have to record separate statistics per gametype.

I agree. But generally, a server hosts only one gametype.

[xlr8or]

XLRstats has event/action bonusses. You can award an event/action such as Capturing a Flag, or Defusing a Bomb seperately in the xlrstats config xml file when they are logged/dispatched by the engine:

Code: xml

	
		
		0.0 
	

It's a fixed addition to the players skill.

Also, XLRstats takes kill-assists into account aswell.

[GedankenBlitz]

Hurr durr.

Since the new version of World of Padman has been released and the new parser is currently in the works, here are some notes.
I think I covered all the data which I initially mentioned.
The only thing missing is verbose damage loglines as well as accuracy. Currently only kills and deaths are logged. Give me some input on how to realize this

My detailed post on the new format mentions the AddScore logline once more. I really think it should be used as an replacement to kills in xlrstats.

[xlr8or]

I like your approach with the addscore addition to the game log. However, for several new games more and more information is stripped from the logs. Leaving us with only kills and deaths to calculate with. Actions such as capturing a flag are removed from the logs... Blame it on the console developers?

How many WOP servers are currently (and previously) using B3? I see no WOP servers in our master list. Where are they?