Gaming Zone > Unreal Tournament 2004

A Bit of Better Balancing

<< < (15/26) > >>

Gil-galad:
 To translate the metric to vctf on this server, there's some important differences to consider :
 
--- Quote from: FAQ ---Games will be played in 3vs3 and 4vs4 teams which makes the whole thing a lot more competitive than most public servers [...] The games will optimally not have any mid-game leavers or trolls. Everyone is supposed to play their best game
--- End quote ---

So, we have matches well over those player counts, and players that join and leave mid game. Piglet mentioned this earlier as a point to  watch out since a balancer that would work for their server could well not work because of these differences.

What I'd try is to further fine tune this logic, by multiplying the ranking/rating points that this system would give you
- in full (by 1) if you played all the match
- in part only if you've only played a part of it. Like : if the game lasts 30 minutes, and you played only 10 minutes, the ranking/rating points you would receive (winning or loosing) should be multiplied by 10/30 (actually, only the +6 part of the points should be multiplied, since the rest would depend on specific actions done while partaking the match  EDIT : not true if every player from the winning receive the same ranking points, like the algorithm used in Assault).

Like this, we could also very well take team switching into account, elaborating a bit more on the above, on top of weighting correctly the value of a partial match. We also have the rejoins to consider : the  time played in a match should not be lost even if a player leaves and rejoins the server, or goes in spectator mode.


Just a small idea, and not sufficient of course to solve all this translation thing, but it's a step.

Gil-galad:
The way I ended up trying to translate their value :

If a win=6,
then one flag capture is worth 6/5, since we need 5 flag captures.
Let's cut that in half since you need to control the 2 flags in order to capture a flag, as an objective (?) way to value a full distance flag run or a full distance flag return, and we obtain 6/5/2=0,6 rating/ranking points for those actions : a full distance flag run with capture or a full distance run and flag return.
But since you have to travel often double the distance to capture a flag than to retrieve it, it would be more adequate to give to a flag capture relatively more points than the points given to a flag return. My estimate would be then be 6/5/3=0,4 for a full flag return, and 0,8 for a full flag capture, because of distances of travel.
 
The most correct way to find a value adapted for a given map map would be to weight the full maximum distance of a flag capture against the same for a flag return, and use the ratio to find the correct divisors to further break the 6/5 that were devised earlier, for the full distance flag capture and the full distance flag retrieval value in ranking/rating points. I proposed a way to measure those distances, - that can probably certainly be improved.

This idea has the following limitation in its conception :
It assumes all the value of the victory is in the flags management, while defending or supporting are in many cases amongst the key elements of what end up enabling the flag captures or the flag retrievals that lead to victory. So there's certainly room for improvement there.
We could probably lower the values again, if we could quantify the other roles/actions that are responsible for a victory.
Why would it be lower? Because the more roles you acknowledge and value, the more distributed the victory's value will be. Perhaps it's obvious that is the chosen logic here, but it's also better plainly written.
One obvious case that pops to mind is the manta/raptor taxying, scorp /bender taxi, ... : if driven or flown by someone, the full value one player on foot would receive would be divided by the number of occupants of the vehicle... or divided between just the driver and the flag carrier, so by 2. Or something better you'll come up with to distribute truthfully the value of the action.

Gil-galad:
One other option, much simpler :


--- Quote from: FAQ ---The rating points awarded to players are calculated as follows:

Points=6+2×(WinnerObjectiveCount−LoserObjectiveCount)
--- End quote ---

We could write it

Points=6+2x(WinnerfFlagCaptureCount−LoserFlagCaptureCount)

Piglet:
You could if you wanted to ignore everything but the number of caps each team got.

That ignores

The number of captures this player got (rather than some good teammate they played with)
The number of times they grabbed the flag
The number of times they helped the flag carrier
The number of times they returned the flag
The number of times they killed the enemy carrying the flag
The points they got
The frags they did
The times they were driving the flag carrier (this isn't currently tracked)
etc...

Gil-galad:
I know. I've got plenty of counts that I'm not considering, but...

Has it been tried?

And has it been tried with the time count correction like I pointed out earlier, to at least avoid some of the noise of the midgame joins/quits/spectates/rejoins/switches?

I'm really curious.

If it's very simple, perhaps it could be tried and trained a bit on some recorded match results. If the algorithm really sort out players efficiently, it should at least be efficient to rank the regular players, no?

Navigation

[0] Message Index

[#] Next page

[*] Previous page