Vicious Infected Versus - TrueSkill Experiment Poll

Read everything below.

Hey everyone,

I've just finished programming the TrueSkill mod! What is it and how does impact everyone?
TrueSkill is an advanced Bayesian statistical algorithm that is designed to measure a player's skill when in combination with a specific team against another team. It is perfect for ranked competitive gameplay in a variety of situations. Some famous games that use this algorithm to rank players and balance teams are Halo 3 and Gears of War (the whole series).

I've created a mod that brings the same ranking system to Left 4 Dead 2. TrueSkill measures the player's skill to a very great degree of certainty. My mod will rank your skill to a value between 0-50, with 50 being supreme (a pro basically) and 0 being a complete noob. Average players are expected to rank between skill 20-35. The mod learns from your gameplay and your true skill becomes more and more precise as you play.

TrueSkill will be implemented in Points Reloaded, but to be completely honest, any statistical data I get from there isn't accurate enough to tweak constants. I need to determine the average standard deviation in a much more competitive server like Vicious Infected. TrueSkill will not hinder any of your gameplay or provide any RPG-like features. It will simply announce a player's skill when they join, and keep track of that skill as they play.

Basically, each player's personal goal is to improve their skill and obtain the highest skill possible-- 50. Who wouldn't want to be a boss at L4D2? However, to increase your skill, you must work together as a team. TrueSkill enforces teamwork, and a lack of teamwork will lower the entire team's skill after a losing match. The trueskill algorithm only checks for two things: 1) did your team win? 2) who were you versing? It doesn't care about the number of headshots you got, the number of infected you killed, the time you completed the round in, or anything like that. It will check to see if your team won and then will consequently increase or decrease players' skills using a very advanced mathematical Gaussian-Bayesian statistical algorithm. It's very effective in determining your true L4D2 skill.

If you think you're a seasoned L4D2 player, then after around 50 or so rounds, you will see your skill ascending to 50. If you simply stink at L4D2, you'll average out in the low 20's. (You must be HORRENDOUS to rank in the 10's lol.)

TrueSkill only measures a player's skill. It does not affect any of the gameplay. VI will still be VI, with the addition of a skill measuring system. That way, you will know who a seasoned L4D2 player is vs a newbie. Skill is all about teamwork, as I have aforementioned. If your team loses, everybody pays the price as the losing team's skill goes down and the winning team's skill goes up, variably depending on the summative skills of each team. It's a very effective algorithm designed at the Microsoft Research facility.

Now, with TrueSkill, you're not only playing to win, but you're playing to work together to improve your skill. As I have said before, I need to obtain some important statistical data to come up with accurate base constants suitable for a game like L4D2. The best server to obtain this data is a competitive one like VI. Will you VI players allow me to install my TrueSkill mod to obtain important statistical data?

If you guys will let me, I need a list of veteran VI players who are very good at L4D2. They show teamwork and almost always win. This way, I can adjust values and see how their sigma (standard deviation) and mu (mean) values change, since I would expect their skills to average out at 50.

So, the poll, "Will you allow me to install TrueSkill to VI? Read the WHOLE description first."

And throw out some good players for me as well as any comments you may have.



Thanks,

Luke

 

Where's the poll? Lol

the VI admins are good examples of excellent players and they could probably name every veteran worth mentioning. But if it's supposed to be a PR system in the end, how does calculating stats on a completely different mod from PR help you adapt it to PR?

Also if I read right the main factors to increasing your TrueSkill level is by winning and who you were up against... One problem on PR VS is the kicking of lowlevels who want to try being a tank. Won't something like this only influence more on the desposal of newbies since the higher levels want the tank to win? Or is it the newbie's job to pass the tank or face the votekick lol

 

Poll's at the top

Right, the constants I'm trying to find are L4D2 related, not PR related. PR has a lot of problems with votekicking and such right now, another reason I don't want to grab any stats from there. The stats I obtain will be adapted to L4D2, and thus will be usable on every server. If the people of VI like the concept of true skill, then I can always leave the plugin running there permanently in a separate database than PR.

 

Is there any kind of punishment to someones rank, if they were to disconnect just before their team looses, for example? If not, might I suggest that the server remembers everyone who was on a team during a round and applies the rank change regardless of whether they are still in game (or no rank change for a win and a negative impact for a loss).

 

Yeah... it's there now lol but I won't vote just yet.

PR is what makes l4d2 more than l4d2. It defines new laws of l4d2 gameplay, the kind of stuff you don't see anywhere else . If this experimentation will really help in the end (it is an AWESOME idea) then I don't see why not however I just hope it won't cause more problems for the lowlevels on PR VS, where level and upgrades mean more than actual teamwork which is the opposite of VI where basic l4d2 skill and teamwork is required.

Good job finishing the mod

 

Chocobo
Doctor Calculator
Zux
Youngblood
Madballs
Antony
Clubroach
Zsoka
)o(G-Force
Sycorax
DreamofDevil
Seeker Sheol
Azurekite
The Hood
Joker
Agent Secret
Cent
Paranosc
hILLO
Wasse
Vulcan
Halcyon
Bonekeep

 

Edit: Thanks for the awesome list Zsoka! You're awesome!

@stefe - TrueSkill is an extremely mathematically advanced ranking and rating system to actually show a player's real skill level, not just the amount of "experience" (server gameplay time) or "levels" earned. This way, matches can be more fair if a player is actually good at the game and not just new to a mod like PR or PRev. You can read more about it here:http://research.microsoft.com/en-us/projects/trueskill/

The really great thing about having my TrueSkill mod in action would be the still incomplete PRev autobalancer. Imagine very evenly balanced competitive teams (the autobalancer disables teamswitching), allowing for fair action-paced gameplay.

The is actually designed to aid low-level PR players. See this post: http://mgftw.com/showthread.php/6692-Level-225-capability-ETA?p=60880#post60880 (for those of you that don't know, PRev is a mod that l1tehawk started and I finished)

 

Yeah, I can vouch for Halcyon on that list. I got votekicked by that dude the other day for being a noob and running back to save someone alone lol. He's a very competitive player. I wonder where he would rank on the system I would imagine 45 to 50.

 

I like this idea however I think it will promote players just jumping to the winning team. If you get points for being on the winning team then players 2will jump more often. Like I said before. Leave VI alone and make another server that is experimental and you can dick around with it anyway you wish. Please leave VI alone.

 

OMFG thank god , finally anti switchers . Nice mod btw.

(talking about pr , i have just played once vi =p.)

 

I started on vi so i know my way around both mods. Lets test it like I mentioned in the pm I sent you luke.

 

That's the problem. I need a live server that is well played and highly competitive. I can't run this on a test server because then the data I get will be filled with a lot of junk data (e.g. bots) and not enough from real humans.

Vulcan, I assure you that VI itself won't be modified. In fact, everything will remain the same, there will be nothing to fear. The only difference is that when a player joins, their TrueSkill is displayed. That's it. Everything else happens in the background and nothing else changes.

As for PR, there are a lot of new upgrades, skills, and features coming out that will utilize TrueSkill to the full extent.

 

No more teamstackers? O.o

 

For skill calculation to be accurate, I will need to block teamswitching and teamstacking. However, if I do that, it will slightly hurt gameplay (sometimes people want to switch teams and join a team filled w/ people instead of just survivor bots).

Instead, I was thinking about allowing teamswitching and teamstacking (maybe), but only calculating trueskill when the server is nearly full (18/20 or higher). That way, there can be no cheaters that decide to pwn on half-empty servers.

 

Sounds interesting. Only way to find out is to give it a try. I too am concerned that once people figure out what they are being scored on they will want to jump to the winning team. Will this encourage team stacking? If I am on a losing team during a round and switch to the winning team right before it ends will I get credit for the winning team? Again, best way to find out is to load it up. Just let us know when and we'll be glad to give you some feedback.

Here are some more names of regulars that Zsoka might have missed:
Ash Housewares
arg77

 

No, there will be no teamstacking if I program it to calculate only if the number of players on the both sides is greater than 9 (thus must have more than 18/20 people in the server). If someone team switches and the teams become unbalanced, trueskill won't be calculated.

Besides, TrueSkill is much more efficient when there are more humans on the server.

By the way, the algorithm considers a bot to be an average player with a TrueSkill of 25.

 

Sure, why not.

I was a bit skeptical about the algorithm at first, but now that I've grasped a better (but still limited) understanding of how it works, I think it might actually work out.

Vulcan made a valid point when he said it might lead to people switching just to increase their TrueSkill. Apart from possible workarounds for that problem (e.g. don't touch somebodys TrueSkill if he switched in the same round. That would be a bit more tricky if you decide which team won by taking the total score instead of the round score, because then you'd have to do that for the whole campaign, which might be too much of a penalty), the algorithm takes the other team's TrueSkill into account.

That means if you beat a high-skilled team, your TrueSkill increases pretty fast, whereas it just slowly increases for winning against noobs. Just like everywhere - high risk, high reward.

It might even make things more interesting and competetive, because if your TrueSkill is pretty high already, you gotta beat lotsa noobs to get better, so you might want to fight high-skills instead of the usual stacking.

 

Yay! A nice l4d2 mod that is based on "TrueSkill"

Thank you so much for this, and I already know where I'm going to land in. Starts with a 5, and ends in a 0

 

I'll briefly explain the algorithm.

There are several variables to keep track of:
Epsilon- draw margin. Basically, how easy it is for the two teams to end up getting a draw.
Beta- usually half of the base mu.
Mu- Mean average.
Sigma- standard deviation.
Round scores- used to determine winner/loser/draw
Psi - cumulative distribution function
Phi - probability density function

By measuring the summation of the sigmas and mus of the team, an accurate delta value for the losing and winning teams can be determined.

However, the base constants (because mu is just m_final - mu_0, where mu_0 is the base) must be accurate to get good data. Think about it like this: the reason I asked for a list of GOOD L4D2 VI players is because "good" is relative and thus I'll need statistical data to set the base values so that you need to be either as good as or better than the base value players to get to a 50.

In other words, let's say I was good enough to get a 50 (lmfao LOL ROFL sure.) Put me in a 1 vs 1 game with any of the following people (who you guys say should all rank 50).

1. Stoned
2. Doctor Calculator
3. Zux
4. Youngblood
5. Madballs
6. Antony
7. Clubroach
8. Zsoka
9. )o(G-Force
10. Sycorax
11. DreamofDevil
12. Seeker Sheol
13. Azurekite
14. The Hood
15. Joker
16. Agent Secret
17. Cent
18. Paranosc
19. hILLO
20. Wasse
21. Vulcan
22. Halcyon
23. Bonekeep
24. Chocobo
25. Vulcan
26. Arg77
27. Ash

Because I'm versing a skill 50, I should be evenly matched. However, I have not yet reached skill 50. The mod has no way of knowing whether or not I am good enough to become skill 50 (probability distribution function). Therefor, if I beat several 50's when I start off, my skill level would rapidly climb the first few games because my sigma value is huge and my mu is low.

Now consider that I'm average, and my trueskill is really around 30. My sigma value is huge, but because I will lose against a skill 50, it's much harder for me to win, and thus the distribution will increase my delta mu only slightly.

This is why we take the summative total of the sigmas and mus per team-- the total will give an accurate depiction of how good or bad the team is, and thus negate or add a suitable delta value.

If we have a really sucky team versing a really good one, and the sucky team loses, their trueskill goes down only very slightly. However, if we had two sucky teams battling each other out in VI, the winner will go up by a lot because both teams are evenly matched.

Thus, the team you are in doesn't matter, but it's always helpful to be on a winning team