[et:qw]: idea: ranks based on XP divided by playtime

(San_Pedro) #21

Hopefully you are spot on. The XP system in ET was/is perfectly fine. I’m sure XP save will be added (many players actually like playing with the rewards they win… I like both non xp save and xp save so i’m good either way)

(Zaedyn) #22

OK, now we’re talking. This is the kind of thing I was hoping for. What I would say is that I am still calculating a player’s rating, but I am not taking into account some effects that I could.

For more interesting info about the pros and cons of systems like mine, see the FAQ on the Trueskill site. It is VERY informative and discusses A LOT of these same issues.

I have considered modeling how a players kill rating affects their player rating. This would model it separate for each player so that I would know how much a player’s killing ability actually contributes to their winning. Some players kill very well, but don’t help their team much and therefore should not be ranked high if winning is the priority. Others kill well and win well.

It’s very hard, however, to tell if you “missed” a shot at a particular player. You have to make assumptions about aiming that may or may not always be true. But I have been considering this as well for enhancing my kill rating model.

In addition, my newly developing kill rating model already takes into account both class and weapon, like you have mentioned. This model hasn’t been published yet because the improvements to overall prediction were not large enough for me to use it. However, I did learn some interesting things about the weapons and classes in general by observing the results of fitting the model. For instance, it is NOT easy to have a high K/D if you use a Panzerfaust. At least, not as easy as everyone usually thinks. By everyone, of course, I mean n00bs.

It is also difficult to model whether players work on objectives. For instance, an engineer in fueldump can go back and do useless things that are important initially, but not later on (rebuilding the bridge, etc). These types of things would be fixable with better game scripting in the scripting engine. I did in fact test a model that took into account the time at which objectives were accomplished. This let me see interesting effects in maps like goldrush where if the Allies at any time got the gold on the truck, they suddenly became very likely to win. In the end, though, these additions did not seriously improve performance.

As I see it, your ideas are good and fun to approximate, and they enhance the initial, simpler model. But to me the simpler model is still a player-based rating. It builds the player rating based on peformance in teams. It even weights the player’s contribution by how long that player stays on each team and tracks early quitting, etc.

Sorry if I laugh at the phrase “dynamic constant”. The system I am using is an improved version of ELO. If I were to use it in a ladder with players playing on the same teams, my system would give the exact same ratings for the teams that a perfect ELO system would give. By perfect I actually mean one where the traditional constant “K” is determined CORRECTLY. Which means by estimating uncertainty in the player ratings and using that to determine K. That is theoretically how it SHOULD be done. That is how my method, Trueskill™, and Mark Glickman’s method do it.

Glickman’s method replaced the original ELO method which used different K-values depending on player skill. So Glickman’s method replaced ELO for the USCF. My method is a generalization of Glicko 1 and 2 to players within teams instead of teams or 1-1 players only.

Now, as for why these ratings are still beneficial, you would have to study up on how Bayesian inference works. It can find what the PLAYER ratings have to be in order for the TEAMS to play as they do over time. This is indeed built for pub-style situations where players move around A LOT. So, in this random mix, you end up finding out who overall wins a lot in the crunch.

I don’t include as much as I could because in the end I just want to focus on winning, I want to only take into account things the player may have control over and exploit. I don’t take into account class because I’m saying, “If a player is really good, he’ll KNOW which class to play, and he WILL play it to help the team”, etc. Again, if you read through some of the trueskill arguments, you’ll see the same thing.

One last comment though. If the same players always play together, and are ALWAYS in the game (not substituted out, etc), then my system and a team-based system will be exactly the same. However, if there ARE substitutions, my system will give higher ratings to players that play LONGER and win, and lower to those who lose. It can get more information about those players. If you read the Trueskill FAQ, pay close attention to the arguments about # of games you need to play and the # of BITS of information. It’s fascinating.

Thanks for the fun discussion.

(extraordinary) #23

I only have a minute to make this reply, so I’ll make it quick then elaborate later…

Team based rating systems are suppose to rate teams as a whole and thus, all players on the same team should have the same rating but in your system, no player should ever have the same rating but they can get infinitesimally close but never be identical yet it would report two players as having the same rating as long as they played in matches together.

When applying the ELO system to a TEAM competition, you’re counting the team as one entity, not as the number of participants on the team and thus, they have no individual rating because they only count as a team.

Thus, your system would be reporting incorrect values for a competitive team than for a pub team. And I still say it averages out the play time, wins and loss of a player the more data you get on that player, thus working in a pub mode.

(B0rsuk) #24

Original poster:
Not as smart idea as you’d think. It would work more less like Accuracy statistics on servers.

You fire 400 bullets and 300 of them hit. Accuracy 75 percent.
etplayer entered the game. fires 2 shots and hits 2ce. Accuracy 100 percent.

(Dr_Tenma) #25

[/quote]

I haven’t read all the posts in this thread but I generally agree with this statement. It seems like a huge number of you guys are confusing the “military rank,” which is supposed to be nothing but an in-game title that indicates your estimated in-game skill and consistent team contributions.

EXP unlocks are the stuff that actually give you added abilities/bonuses to gameplay and these reset at the end of each campaign. Period. They’ve never been announced to be “persistent” and this won’t likely change.

The system sounds perfect to me, I don’t know why there’s so much debate. It won’t be imbalanced because hardcore gamers have all the unlocks, all the time. And it’s not like you have to really respect or take commands from someone with a high ranking title. On World of Warcraft’s old honor system, I saw a pretty good number of guys ranked Marshall/Warlord or GM/HWL and they weren’t necessarily all that good (obviously just played a hell of a lot). You probably don’t need to pay attention to that rank at all, if you don’t want to.

(Ragnar_40k) #26

We had this system on splatterladder.com but dropped it for an ELO approach, since the gained XP is different for each mod/map/server and you end up with players playing on certain servers with certain maps and mods only to exploit the ranking system. :roll: