Performance and elo calculation

[ichwillrotsein]

I noticed that the performance shown in the game (endgame screen) differs from the performance you use to appraise the computed elo change. I mostly have the highest performance points in the game but when I check my Δ on league.btanks.net, sometimes guys with 100 points diffenerence have a better factor, like 1.0 for them and 0.95 for me. And those people didnt even contribute very much for the win (in my eyes ). I guess the performance in game focuses the teamplay more and the Δ is rather based on hero kills, creep kills and building kills.
I'm afraid I have no example at the moment because I forgot to take screenshots.

[BENNIE.FM]

Hrmm interesting topic well i honestly think its nearly almost always accurate with a few exceptions maybe a few tweaks could be made here and there but i think they have a done a good job.

One think i noticed elo does not take into account factors such as sometimes i vs 1500 elo me 1600 elo but they hold me for a long time and i punish them and it ends that the lane player against me is then punished and are penalized most when they loose,but elo does not take into account the ability of me or how well he did to hold me of for so long for example when no one his team could of done better.Elo sometimes screws someone who has the hardest jobs often happens with support play.

It just doesn't show which kills your assisted in some players are usually harder to kill or the kill it self turned the game etc.If you assist in killing a weak player not much but when you ganking or killing a better player its more vital to the end result of the game.

I don't know if the elo takes into account the amount of times you defended a cp sometimes some good players have no ability to see a cp battle coming or defend.But still get 0.15 or 1.00 because of other statistics even though defending cps is just as important.

Sometimes playing guard gives you no kills etc or not much creeps but assit the team alot .Even deaths defending a cp alone or team mates not helping.Even orbital gives large aoe damage item usually played with hunter but often the other players are just standing there pick up the kills.Just a few examples

Btw these are just a few thoughts im sure there are many others i which will come up.

I know what you mean though it happens however.I don't know if any of my comments can give way to some suggestions.How about having a little more weighing on other factors im not sure just a suggestion i am quite happy with the elo system

[Exodus]

You are right, the performance value in the game is a different one, compared to the one on the league page.

The league performance takes nearly every tracked statistic into account and weights them according to some predefined values.

The ingame performance however calculates two values. The first one assigns a value to each team. This value is the sum of all kills, deaths and support points of a team, which are multiplied by their weights respectively (those weights are the same as the league weights for kills, deaths and supports). This first value rates the teams against each other and since every player contributed to the team result, they all get this value as a base.
The second value is the earned gold. Each player gets a factor that shows how much gold he earned compared to his teammates. This factor then gets multiplied with the teamvalue and results in the individual values.

I used the gold factor instead of every single stat, because I thought this would be more stable and in the end, you do everything to earn more gold and to get an advantage. So I thought it was fitting.


As you can see, you were more or less also right about the ingame performance value being more about teamplay than the one from the league page.




Also, if you have some time or you are interested in this, it could be helpful if you compare some performance results from the game and league page and tell us, which you think fits better (while also showing the numbers, so we are able to follow your reasoning). This might help us with future adjustments.

[ichwillrotsein]

Also, if you have some time or you are interested in this, it could be helpful if you compare some performance results from the game and league page and tell us, which you think fits better (while also showing the numbers, so we are able to follow your reasoning). This might help us with future adjustments.

My first game after my first post is already perfectly suitable

ip = ingame performance
lp = leage performance
and the "," is the "." in your world

My stats:
ip: 1,528 lp: 0,92

Radaschille's stats:
ip: 1,336 lp: 1,00

mocuishe's stats:
ip: 1,173 lp: 0,94

So, although I have 14%/30% more points then Rada/mocu "ingame",
there factor for the leage ranking is 53%/13% higher then mine (based on 0,85 as 0% and 1,00 as 100%)

To proof it, this time i have screenshots
Ingame victory screen    
League scoring     ( wich is also here: http://league.btanks.net/game.php?gameid=36548 )

Well thats the data survey, I will sign again when I though about it ...

@Exodus: Would be nice if you could post (or pm me) the formula for ingame performance and league performance

Btw I made the game, this Radaschille guy was just a mid/longrange/airshit/last_hit_shocker noob no offense

If you're interested in the game, here's also the replay. The game itself wasnt that gr8, we had two bot related disconnects, both dark foce ...

---

And please remove this annoying 15 minutes time cap for editing posts

[BENNIE.FM]

I have noticed this as well.Nice obs but Exo do you need more replays or comparisons.

[eSVau]

As Exodus stated both performance values are differently calculated. I think the ingame one is better but the league does not have those gold information and derives the performance from the other values.

[Exodus]

I have noticed this as well.Nice obs but Exo do you need more replays or comparisons.

Sure, the more the merrier.

[stibi-]

ban grizzly for being afk ...

[ichwillrotsein]

I think the ingame one is better but the league does not have those gold information

The league has informations about the total hero kill bounty, creep kills, building kills and cp captures. Those values are simular to the earned gold ingame.
Maybe one creep or building kill is more worth in the ingame performance, or you assess something twice by rating the total hero kills plus the hero kill bounty (wich is the same if you neglect the time in game where those kills were made) .... wich brings me to my next point:
The stats I uploaded show, that i earned the most gold, killed the most creeps and buildings, have the highest lvl, most CPTPs and lowest deaths.
The other guys are better in tank worth (cause i spend around 6500 gold for facs in the end to finish base), kills (long range airship last hit shocker ), total dmg dealt (long range ... ) and assists, support (btw what is support?)
So the final tank worth at the moment when the enemy castle falls is very useless to evaluate the performance factor, because some people spended much money for armor upgrades, factorys, breakers, barricades, or even money loss due to a fast change to inf .... and so on, the list here is long. Well and some people dont spend gold for the team ...

A good formula would be for light force (lf):

Xi = [a * (personal hero kills / total hero kills lf)]
+ [b * (personal creep kill / total creep kills lf)]
+ [c * (personal building kills / total building kills lf)]
+ [d * (personal tank lvl / 25) - (25 / average tank lvl lf)]
+ [e * (personal CPTPs / total CPTPs lf)]
+ [f * (personal assist / total assist lf)]
+ [g * (personal support / total support lf)]
+ [h * (personal CPs captured / average CPs captured)]

Δi = 0,85 + 0,15 * (Xi* / Xi*(max))

with:
Xi* = Xi - Xi(min)
Xi*(max) = Xi(max) - Xi(min)
and (for example) {a,b = 0,2} {c,d,e,f,g,h = 0,1}

It's easier with excel

the actual weighting of those factors would be set based on empiric values.

[BENNIE.FM]

  LastReplay.w3g (Size: 2.98 MB / Downloads: 227) http://league.btanks.net/game.php?gameid=36565

here is another game similar to what happened in ichwillrotseins game but i ended up getting the 1.0 Another values varied from in game performance im not sure how to upload a screen shot of the actual end game performance but will make some comments when i have some time.

[Max]

Here is another example : http://league.btanks.net/game.php?gameid=36575
Carizma got almost 2 times more points than me in the end of the game and i get almost same as Forex but i got 1.00 in the end, Carizma 0.99 and Forex 0.92.

[RAF_Today_Hero]

omg

[urnightmarre]

ban grizzly for being afk ...

MAN U OUT OF SUBJECT

[GEN_Schwarzkopf]

I think there are many intangible statistics that lead to victory, particularly in supporting roles. For certain helping take a CP whether you lived or died should receive credit. Helping capture a CP successfully should be given huge rewards, and building kills should be much less.

Kill assists can be a very big factor, and especially undervalued for support players. In most cases the stun tank does not finish the kill but the kill would not be possible in the first place without it.

When it comes to deaths, dying because you're the only person close enough to tp in to save a CP in time should be rewarded. There so often are people chasing player or tower kills the responsible players die to save the rest of the team from suffering. I constantly do this, as well as share damage to save players, but again this is an intangible statistic.

Performance needs to weigh less on the farming and more on the assists and support.