next »

[doctor_x]

I think there is a bug in the skill point calculation that was released about a month ago. The skill point stats on UrT tdm SNT #3 Uptown are really getting inflated. I was #3 on the server and didn't play for a few weeks. I dropped to #15 without playing at all, while seeing the skill points of the other players skyrocket.

It seems that the new version gives you points for assist kills. Do you remember to subtract the same number of skill points from the victim? If not, it would explain the skill point inflation.

[xlr8or]

Assistskill mechanism is brand new, bleeding edge. So far we've been compensating assistskill partially for the killer, but not the victim. What it does basically is this:

1.) assister shoots the victim to be, the assist is tagged on the assister with a timelabel.
2.) killer shoots the victim within 2 seconds.
3.) the assisters win probability is calculated (could be multiple assisters), awarded to the assisters and summed up to a total
4.) the killer and victims win probabilities are calculated and the victims skill loss is deducted.
5.) the (sum of) assistskill is deducted from the killers skill addition, to a max of 50%, so the killer will at least win 50% of his calculated skill for making the kill.

This, together with a minor bugfix will indeed inflate total skill gained per player. The issue now is that I'm waiting to see where the skill will balance out. We'll have to reduce the kill-bonus to reduce the current inflation. But it needs a little time to get the settings fine-tuned.

[doctor_x]

Thanks for clearing it up. So... is the total rating gained equal to the total rating lost? As I understand Elo rating, that should be the case (assuming you're using some form of Elo rating)
Just curious.

[xlr8or]

Well, you made me think and you put me on the right track, so I'm in the middle of modifying the code somewhat.

The victim now also suffers from skill loss from an identified assist. This whole thing is pretty complex, but doable. Because the assist can only be positively identified when the actual kill is made, not sooner. Also we have to deal with different skilled players, so all ELO calculated win probabilities should be taken into consideration when calculating skill for killer, victim and assister.

Now it flows like this (this is hard to explain, so ask if you don't get it):

1.) assister shoots the victim to be, the assist is tagged on the assister with a timelabel.
2.) killer shoots the victim within 2 seconds.
3.) the assisters and victims win probabilities are calculated (could be multiple assisters, assistbonus of only 50% taken into account), awarded to the assisters, deducted from the victim and summed up to totals
4.) the killer and victims win probabilities are calculated, assists are already settled.
5.) the (sum of) assistskill is deducted from the killers skill addition, to a max of 50%, so the killer will at least win 50% of his calculated skill for making the kill.
6.) the (sum of) assistskill deductions is reduced from the victims skill-loss up to a max of 50%, since the victim was already punished for skill loss from assists.

This way all assist, kill and victims skill is taken into account to a certain extent. The 50% is to make sure that at least half of all skill calculation weighs on the kill, not on the assists. The assistbonus is 0.5 x normal calculation (without the kill bonus) for an assister, so that reduces weight for an assist also by 50%. The total skill gained versus the total skill lost is therefore not exaclty the same, since we weigh assists different from kills. In the original version (prior to the assist calculations) we also had to compensate for loosing skillpoints from the general skill pool (suicide penalties and teamkill penalties). We did that using the kill_bonus.

If it where just you and me on the server and we had no penalties, we'd always have a grand total of skill points that is the same. With a large player crowd every player starts and brings 1000 points to the table. Suicide penalties and teamkills make skill flow away from the skill pool. A good player that never returns keeps skill points that become unreachable. Also the weapon and action bonuses influence the skill pool. My aim is to ensure a somewhat positive, but well balanced skill pool. Where new players are still able to reach the top, and the top 10 is always challenged.

The skill bonus should be reduced when players become out of reach for starters and should be raised when average skill drops below 1000. This is a per server setting because games play totally different. On a tactical CoD server with a lot of returning regulars the pool will develop different from a UrT server with many one time players.

Thanks for thinking along with me.

[doctor_x]

Hey, just glad to help. I don't want to see the skill points drift to infinity on my favorite server