[Today] Tremstats Version 0.5

[squishy]

Thanks, DASPRiD. That's helpful, but let me see if I get this straight.

You are renaming the games.log to games.log.tremstats.tmp. Tremulous will continue to write to this TMP file until a game completes. Tremulous then creates a new games.log. Now TremStats will process the tmp file?

I assume this is a non-Windows tremded feature.

The only option I could work out in Windows would be to add a check for the last instance of ShutdownGame, and ignore the rest until next time around.

[DASPRiD]

You got it, that's the way I work

By the way, if you got a running workaround for windows, I may implement it in the svn-version. Just send me the patch.

If you do so, please checkout the latest version from SVN:

http://svn.dasprids.de/repos/tremstats/trunk/

[squishy]

Finally got things working. First I copy the log to the archive directory. Then I strip unwanted ending lines from the log (everything after the last ShutdownGame) - this removes unfinished games. Then we compare the file with the "last version" and process new games.

I'll try to work my method into the latest SVN and post for your review.

May need to create a variable in config for the user to tell whether they are using a Windows server or not.

[DASPRiD]

No need to:

Code: [Select]


import sys

if sys.platform == 'win32':
  print "I am windoze."
else:
  print "I am some good OS."


And btw, wht exacly do you do with the stripped line? Do you drop them?

[squishy]

They are dropped. They would get picked up the next time around.

[DASPRiD]

Ah ok, then it's fine.

[squishy]

Just pulled down the SVN and input my changes. Here is a ZIP file including the changes and a small changelog noting the lines I updated.

Feel free to contact me via ICQ or PM to discuss...

//**URL REMOVED**//

//I also added a tally of "says" for the players table, so I could display Most Talkative on my stats page. This change is represented in structure.sql and data_calc.py

[DASPRiD]

Alright, I'm going to contact you per ICQ.

[rasz_pl]

I think that "time factor" is wrong and should be abandoned. Its a buggy patch over a bug. It makes average server campers look better than good players with real life .

 Better way is to sort ppl according to the kill effeciency, and then (multiply kill effeciency of ppl player killed[or number from the sort above] x times he killed them)/ total kills, and then sort on this number. ppl with 2-5 games played usually have low chance to meet good players, and if they do they dont make >10 kill effeciency , and if they DO then they are really good (they managed 10 kill effeciency against other good players).

That would factor out the time spend on the server, and really compare players against each other. Killing feeder n00bs would lose importance.
Of course it would be even better if the parser would store class of killed player (now it only stores class/weapon of the attacker), then we could also factor tyrrant pwning naked rifles and dretch killing BSes. Imo it should be in the next version.

PS1:
and just to back up with a strong argument look at

29   5555555555555555555555555]
Game-time factor   0.01
Kills   20
Total Deaths   2
Kill Efficiency   6.67

34   Xonya[FI]
Game-time factor   0.04
Kills   395
Total Deaths   228
Kill Efficiency   2.47

I mean LOL, Xonya is a good player, she kills me a lot, and there we have some lamer who happened to play twice pwning n00bs left and right (maybe he joined at S3 and ran arround with camped pulse/lucy after dretches?) 5 places higher than her ... cmon, the main table MUST be relative, each player compared with every other equally. Its not fair to just factor time spend on a server and act like its all good in the hood.

PS2: and no, im not ranting because Oli is about to take first place because he camped bigger time factor than me :] hehe. Actually I didnt even tried my formula yet on the dataset you send me and dont know where I will end up :/

[DASPRiD]

So you want a player with one 1 game played and 100 kills / 0 death against noobs to be on first place? :roll:

[David]

Make it (kills-deaths)/time or whatever, and then do <25 games = not on the list.

[squishy]

//NEVERMIND!

[DASPRiD]

One negative I've found is that player_games_played includes spec-only connections.

It doesn't. If you spec, you get no kills/deaths/teamkills. If you have zero of them overal, you don't get plus one player_games_played.

Make it (kills-deaths)/time or whatever, and then do <25 games = not on the list.

Thats nearly the way I do it, but it's more complex.

[squishy]

I just saw the line in the source. Sorry!

[rasz_pl]

So you want a player with one 1 game played and 100 kills / 0 death against noobs to be on first place? :roll:

OH F... I wrote A4 poem about it and fokin browser refreshed the page, oh CRAP !!!
anyway

(multiply kill effeciency of ppl player killed x times he killed them)/ total kills

so for your player, lets say he killed 100 times this n00b (sorry Tric )
http://tremstats.dasprids.de/player_details.php?player_id=2161

0.86 x 100/100 = 0.86, he is as "good" as the ppl he killed. Now I see that my equation is missing the second part

(multiply kill effeciency of ppl that killed player x times they killed him)/ total deaths

so we have to ignore ppl with ~<10 deaths to not skew the statistic, or ~<10 deaths and above 100 kills.

then we can do (multiply Destruction Efficiency of ppl player killed x times he killed them)/ total kills * player Destruction Efficiency .. or not

 first both numbers are better while bigger, we could multiply them with each other or do some other fancy stuff with them, but essentially I want THOSE numbers to decide, not some artificial "I play here every day so I pwn" factor

Its kinda hard to avoid killing noobs, so its highly unlikly that someone will thrive selectively killing only good players , on the other hand avoiding getting killed is a skill no matter who tries to kill you

hmmm hmm, ok,  good players playing only with good players, this can be a problem. It will make ppl playing with n00bs look bad. There must be something smart, that has nothing to do with how much you camped the server

hmmm maybe we should compute kill efficiency individually for every map played, and THAT number would multiply against
(multiply kill effeciency of ppl player killed x times he killed them)/ total kills on that particular map

 Oh cmon, work with me, throw me a bone here.


ps: one PAINFULL thing is lack of parsing into the database the class/weapon of the player that died (I know it needs patching to have that info). Its one thing to dretch a BS and die 3 times in the process, its the other to tyrr 30 naked rifles. Same goes the other way arround, one thing to die by the tyrr, the other one to get dretched while armored/hatted/shotgunned/and nade loaded

[DASPRiD]

Btw, you totaly forgot the destruction efficiency. Tremulous is not about just killing everyone. Anyway, the game time factor has just another effect: If you stop playing on the server, your efficiency gets down and you get out of the top-stats.

But we could modify the algorithm so regular players don't see a big difference in their gtf.

[rasz_pl]

Btw, you totaly forgot the destruction efficiency. Tremulous is not about just killing everyone.

no I didnt
>then we can do (multiply Destruction Efficiency of ppl player killed x
>times he killed them)/ total kills * player Destruction Efficiency .. or not

But we could modify the algorithm so regular players don't see a big difference in their gtf.

Could you at least make those
(multiply kill effeciency of ppl player killed x times he killed them)/ total kills
(multiply kill effeciency of ppl that killed player x times they killed him)/ total deaths

a part of the equation?

[DASPRiD]

ah, to compare them with eachother... hmmm, sounds like a good addition, i'll test it if i find some time. btw, let's not use the forum as chat, contact me over icq, jabber or irc.

[rasz_pl]

soo wazzup with Tremstats v0.6.0?
what is a Games paused?

anyway I like it (since Oli is no longer a threat )

[vcxzet]

does it list how many tyrants you killed?

[DASPRiD]

rasz_pl:

Games paused is how many games you have paused. There are 2 treshold variables in the config now: games_played and games_paused. The first indicates, with how many games min played you get in the list. The standard for this is 25. the games_paused treshold means, how many games you can max have paused before you get out of the list, standard is 500. If you start playing again, you get immediently back in the list with your old stats.

Also, I found a good use for the Game-time factor: It's used for the most active players, instead of just the games played.

vcxzet:
It's not listed yet, what exacly you killed how often, but this will follow.

[rasz_pl]

rasz_pl:

Games paused is how many games you have paused. There are 2 treshold variables in the config now: games_played and games_paused. The first indicates, with how many games min played you get in the list. The standard for this is 25. the games_paused treshold means, how many games you can max have paused before you get out of the list, standard is 500. If you start playing again, you get immediently back in the list with your old stats.

Also, I found a good use for the Game-time factor: It's used for the most active players, instead of just the games played.

vcxzet:
It's not listed yet, what exacly you killed how often, but this will follow.

1 so if i pause for >xx games i get dropped back? sounds ok
2 Queen of scotland landed higher than me, so we need to tweak the algo once again :]

>ps: one PAINFULL thing is lack of parsing into the database the class/
>weapon of the player that died (I know it needs patching to have that >info). Its one thing to dretch a BS and die 3 times in the process, its the
>other to tyrr 30 naked rifles. Same goes the other way arround, one
>thing to die by the tyrr, the other one to get dretched while armored/
>hatted/shotgunned/and nade loaded

MDing poor dretches with 500-600 credits worth of equipment should not be equal to dretching 500-600 cred humans

if we could only parse the info about class/equipment of the killer and the killed person, we could come up with some weight system. For example the percentage ratio between alien class HP and dps vs human hp * armory class(can be tricky, different armory areas/hit zones) and weapon dps.

[DASPRiD]

And can't repeat often enough, there are no hp's logged, and I can't base the entire kill efficiency on something non-standard. Only with the Client_TeamClassPatch I firstly always know, who's playing which class. Before, I only know, whom you killed with which weapon.

[rasz_pl]

And can't repeat often enough, there are no hp's logged

>I know it needs patching to have that info

[DASPRiD]

Right, and not everyone want's to patch his server, so I can't base the calculation on this.

[rasz_pl]

Right, and not everyone want's to patch his server, so I can't base the calculation on this.

that could be an option for anal numbern00bs like me just add 3 more fields to the KILL table and let admins use normal logs (ignoring those fields) or pathed (using them)
or not, just throwing ideas out loud

edit: actually you can partially kinda almost guess what was the class/weapon of the killed person, just make the parser ass_ume that after the last person the killed one killed he didnt evo/buy, and write to the database his last known evo/weapon