A new factor evaluating the performance (to be added in the player's statistics)

[Leo]

Two factors are quite accurate no matter the hero's role in the game.

(kills+assists)/deaths
This is already implemented. No matter carry or support, this is very accurate. If you have >4, then you are good.  But this only gives info about your individual skill. You might have in a game 5-0-5, but this would be a bad score if your team is winning like 50-5.

(your kills + your assists)/(team kills)
This evaluates your teamplay. It ignores how much you have died, but shows for how many of team kills you have helped.  Under 50% is considered as not a good teamplay.

So maybe to evaluate both teamplay and personal performance at once a possible factor would be:

P = (your kills + your assists)²/(your deaths)(team kills)

P<1 is considered bad.

What do you think?

[Alucardo]

+1

[Bart]

It looks promising, but....

First of all there are 5 different roles in team....... supporters and gankers get most of k + a from total teamkills
while hard-carry is farming 30-50mins to get items and win game for the team so.... invalid imo.

[Leo]

You state that a carry will farm and won't participate in the early stages.
This is true only for very few heroes.
If a player is playing such a hero, he will farm and avoid fights. This means that his deaths will be also low as his assists and kills.
This means that his 1st factor (kills+assists)/deaths won't be different than high rates of assists and deaths of the supporters. Lategame, he can do many kills, to fail or the game to end. In 1st case, he will get a good (kills+assists)/(team kills), in the 2nd he will fail to do so and in 3rd, even his hard farming , his team won the game 4vs5 and he does not deserve a very good valuation.

In contradiction, I think that the people having (K+A)/D higher than 5 are actually playing mostly as carry. 

To summarize: If a carry at the end of the game does not have enough K+A , then he failed, so he deserves his low P, and that is a valid statement.

[Bart]

But sometimes games end in 30mins when "farmer"has stat 1/0/0 for example but he was doing good job at cs etc....how would you evaluate this, it doesnt show his skills neither it will make him look good, in fact it will show he is "noob" in your formula.

[Leo]

But sometimes games end in 30mins when "farmer"has stat 1/0/0 for example but he was doing good job at cs etc....how would you evaluate this, it doesnt show his skills neither it will make him look good, in fact it will show he is "noob" in your formula.

There are two cases of this situation: winning or losing.
1. Winning
If so, then his team has won without him, thus he has been useless as a teamplayer. He could've participated with his team earlier so the game could have ended even earlier. He gets what he deserves - not very good evaluation.

2. Losing
His team kills will be low as his own K+A.  When losing, you cannot afford to fully ignore your team. 

Anyway, I don't say the formula is perfect.
As you say: http://playdota.eu/stats/game/2400861/rdeupd1600
Since Kunka was a whiny kid and intentionally was afk ~20mins watching us struggling 4vs5 and me as Void was aiming to farm and prolong the game as much as possible. At the end Kunka seeing us that we are still not losing even his childish behavior, decided to come back in the game.
Based on my formula, he with his 9-4-15 will have a better P than me with mine 4-1-11. 

So, what tweaks do you suggest to improve the formula?

But such cases are quite rare. It is more common a good player with a non-carry hero to actually to carry the game.

Anyway, even with the flaw regarding the carry heroes, isn't that a good thing? Because almost everyone thinks himself for a pro and pick a carry, and such games turn into big disasters. Let's encourage a bit the support.

[donjacrtasamir]

why don't you just calculate all stats of every player who played let's say more then 100 games (to avoid noobs maphakcer and new accounters) so u would get  the mean value of all players

lets say  mean value of kills =10
mean value of deaths = 5
mean value of assistans = 8
cs = 100
denies= 20.....

and then u can compare your stats with the mean value of every player.

you get +/- points for  better or  worse skills  if u have 20 kills per game u get like +10 points
if u have 15 kills per game you get +8 or 7 points and if u have 3 kills u get - 2 points.(something like this)

lets say a normal player has 50 points the more points you have the better you are.

this would be used only to balance games  and it could be used only on players who played more then 100 games cause .I say more then 100 games because someone could play just 1 game trapemso noobs and get 100 kills and 0 deaths and be the best player on playdota.eu


oh and 

P = (your kills + your assists)²/(your deaths)(team kills)

P<1 is considered bad.

What do you think?

i think the problem is if you have  0 deaths you always get  (your deaths)(team kills)=0 [team kills don't count if your deaths=0 ) so (your kills + your assists) /0 =NAN

[Leo]

P is not calculated over 1 game. It is based on sums of your stats of all your games. Same as the implemented K/D, A/D and (K+A)/D statistics. How many games someone ends with 0 deaths?!
Is that better:
P = (all your kills + all your assists)²/(all your deaths)(all team kills)(played games)

[donjacrtasamir]

P is not calculated over 1 game. It is based on sums of your stats of all your games.

so why don't u use my system.I think there is nothing better then the mean value and comparing something to its mean value.

mean value of kills = (kills of player1 + kills of player2 + kills of player3 + kills of player4 + ...... kills of playern)/n

 compare mean value of kills with your kills and get +/- points

[chelom]

I agree with you but the number of games influence how good you are. A person with high numbers of games played is better than a new player.

I suggest the equation should be something like this

P = {(kills + assists)/(Total kills team)+(sqrt((aver. kill)^2+(aver.assist)^2))/1.414213562*max(avrg.kill, avrg.assist)+(games played)/(games played+10)}/3

So this equations have 3 parts let me explain you those parts

1) With this we will messuare how active is a player.
2) we will measure the difference between kills and deaths. On this equation If both kills and assist have the same value u will get the higher number from that part of the equation. However this can be changed. So if deaths and assist differ from an especific percentage, they will get the higher number of the formula.
 3) we will measure how good a player is regarding how much games he have played. This will reduce users who mass accounts.

Another thing is that the highes value u can get from this formula is 1. That mean u should be very active killing, and keeping your average kills and assist on pair to get close to 1. I think those factors like creepkills and number of death are somehow on the formula. If u have a lot of deaths, the first part will be reduced cause you team will probably kill when u are death. Also late in the game, if u didnt farm enough u will died more easily and wont kill as much. That mean noobs on pro games wont get any better. (One huge problem resolve). I know the formula requieres a lot of work but i think its a good approuch of what u guys want on a formula.

[Leo]

About 2nd part:
Where are included the deaths?
What is max(avrg.kill, avrg.assist) ? Without even multiplying with that, the (sqrt((aver. kill)^2+(aver.assist)^2))/1.41 can go over 10, which totally overrules the 1st and and 3rd parts which are lower than 1.

About 1st and 3rd part:
Part 3 is bigger than Part 1 and overrules it. This is not right. The played games at the least important factor. DotA is not a new game and if someone has ~50 games this doesn't mean that he is weaker than someone with ~500 games.
For a regular player part 1 will be =0.5 and part 3 for 90 played games will be 0.9.

I miss the logic in this a bit.

[chelom]

Well u do bad the part 2, max(x,y) takes the highes value if x>y then it takes x if y>x it takes y and if both are equal takes x.

The deaths are not included cause it affect your score already.

I dont see how u go to 10 with this part

(sqrt((aver. kill)^2+(aver.assist)^2))/1.414213562*max(avrg.kill, avrg.assist)

Well the denominator have a parenthesis ofc.


(sqrt((aver. kill)^2+(aver.assist)^2))/(1.414213562*max(avrg.kill, avrg.assist))

like that.

[Alucardo]

Well u do bad the part 2, max(x,y) takes the highes value if x>y then it takes x if y>x it takes y and if both are equal takes x.

The deaths are not included cause it affect your score already.

I dont see how u go to 10 with this part

(sqrt((aver. kill)^2+(aver.assist)^2))/1.414213562*max(avrg.kill, avrg.assist)

Well the denominator have a parenthesis ofc.


(sqrt((aver. kill)^2+(aver.assist)^2))/(1.414213562*max(avrg.kill, avrg.assist))

like that.

professor chelom lol

[Leo]

How does Deaths already affect the score already?

Part2: What the goal of that? Why a player will be considered more skillful if he has equal K and A rather than more Kills ?
A score 20-2-0 is better than 10-2-10, which is better than 5-10-5, but here the 1st will be the minimum of 1/1.41 and the other two will be the maximum - 1.

Even Part 3 sounds a bit more reasonable - at least to favours the more active players, Part 2 makes no sense at all.
"Hey, I am 10-0-0 during the game. Now I should stop killing the enemies and do some assists instead to gain a better score"

Please, explain.

[chelom]

cause the deaths affect your ability to kill and to do evertyhing in fact. Make you weaker mean less kills and stuff.

The goal of part 2 is avoid players for last hit on heroes cause they need to balance kills and assist not just kills. Of it can be changed the fact that kills and assist have to be the same. I think assist should be like 1.5times bigger than assist so that part gives u 1 but its just the idea.

"Hey, I am 10-0-0 during the game. Now I should stop killing the enemies and do some assists instead to gain a better score"[/i]

And ofc that the idea exactly.  And believe u wont be 10-0-0, another player might . The whole idea of this is to prevent "good players" to leach points from noob games.