Forum

trath for president eu tf2

Not kidding, really great job Trath. I hope to see it working in a near future :)

Is there any way to make “ELO” hidden? Then any player care about it.

If you were going to do it, TrueSkill would be the best system. Microsoft created it for Xbox Live and while it’s a lot more complex than ELO, it’s easily the best way to rate skill in team games. But again, not enough data.

Quoted from Skyride

If you were going to do it, [TrueSkill](http://en.wikipedia.org/wiki/TrueSkill)
[…]

2muchreddit?

Quoted from Skyride

If you were going to do it, TrueSkill would be the best system. Microsoft created it for Xbox Live and while it’s a lot more complex than ELO, it’s easily the best way to rate skill in team games. But again, not enough data.

From experience of what happened in Dawn of War 2, no. Just no. It didn’t work out at all – or they messed things up

*Update/Current Status:
After a lot of thinking, I managed to understand how to theoretically do what I wanted to do. ( I am not a pro coder, so I was never introduced to the thought of a for loop inside of a for loop – now it just seems plain simple and I feel dumb)
I wrote it all down in code, but had an error – finally understood the problem and fixed it – got another error. Things just aren’t working as I want them to. So theoretically “I got dis”, just gonna take time till I get it working for real.

Quoted from Nymthae

Damage on stats would be good, if you can get the heal spreads as well that would be cool.

Those are both stuff that require the Supplemental Stats plugin, which would require SourceMod on server. I’m not actually sure but from what I’ve seen written here and there, that is something that is to be avoided?

Also just in general, I don’t see how heal spread and damage or dpm in a pickup could be remotely interesting to anyone. I think only thing that might interest someone are the stuff I suggested.

Quoted from herpderp

[…]

The only thing that irks me about ELO is that it’s easily skewed, and people put too much weight on the ELO rating (see MGE servers with ELO rating).

Sounds like a good plan though, but I remain skeptical.

“easily skewed” ? I guess some people will take their ELO score to serious, but so what?
Yes, obviously I can “hide” people’s ELO – not put the option of showing their ELO in each class when they type “!rank” in IRC, but I don’t see any good reason why it should be hidden

According to the ELO system, a div 6 team vs a top prem team will lose 999/1000 matches. So if anybody takes ELO serious, now they know – play 1000 matches vs a top prem team and you will be able to say “I won against top prem” for sure!

Quoted from Skyride

If you were going to do it, TrueSkill would be the best system. Microsoft created it for Xbox Live and while it’s a lot more complex than ELO, it’s easily the best way to rate skill in team games. But again, not enough data.

Even though it does seem more complex, I don’t think that system would ever work in any game. I think it goes overboard.
Regardless of what I think, I don’t see how you would know the “trust in rating” for each player, given that you win/lose as a team.

Also whenever I PM you on IRC you seem to either talk a little then suddenly stop responding, or you just don’t respond at all.

Quoted from SiTeHBu0mbb

Id’go with something that monkeh suggested,
that the amount of ‘divs’ added up should be the lowest.

Every div gets a point modifier: prem=0, div6=6. Add all these points together and both teams should be on equal or near equal points. The second selection-set (as to not have teams that consist of 4 prem players and 2 div6 players) could be something along the lines of a div checker; so if the mix consists of 4 prem, 4 div1, a div2, div4 and 2 div6, it automatically splits up even numbers, which would give:
TEAM1: 2 prem, 2 div1, div4 and div6 (=2*0+2*1+4+6=12)
TEAM2: 2 prem, 2div1, div2 and div6 (=2*0+2*1+2+6=10)

You can already see how this is far from idea, so another factor gathered from played matches (average performance) could be implemented later on to balance this better.

A way more simple approach would be (trusting on the scene though!!) just to make the teams with division behind their nickname. If 4 players (for example) dont agree with the teams, they can use a !scramble command or !autobalance (which will have some sort of script behind it)
This last option should reduce the amount of coding dramatically:)

It doesn’t matter if your using “divs” or “ELO score”. divs is “6-0(prem)”, ELO is “600-1800”.
The reason for using ELO over “divs” is that ELO allows for way more “in-between” rating, and there is no reason to change the ELO math to accommodate the way of “divs”, when simply putting divs in ELO is much simpler.

After every pickup, based on who won and the expectancy to win for each team and player, every player’s individual ELO score on the class he played changes accordingly.

Your suggestion of allowing a re-scramble simply wouldn’t work. Not just because of the abuse (people will want to be with a friend or don’t want to be with a certain player, people think that their team sucks ETC), but also because another random scramble of teams isn’t going to come up with more balanced teams. You could randomly scramble teams 3000 times and you still have no guarantee that teams will be close. That is why this has to be done the long/hard/right way. Teams are as even as possible – no exceptions.

Quoted from dodgydogman

Id go with the weight option lots of people are talking about. Ie demo is like 1.5, solly 1, medic 1, scouts 0.8, and times that by their respective divs and get an even as possible team. Obv playtest and tweak.

Maybe have a vote system so people can choose whether to let the medic choose or use the divs system, or even a crude elo, see what most people are using, and focus the most time in fine tuning the most popular system.. Give options to re scramble teams if they dont like the set ones etc.

Are you going to add a way for teams to search for mixes near their skill, or other teams? That would be cool, if that got popular, an elo type system would work really well, though tf2 probably isnt popular enough to actually get any reliable results. It would also probably make finding a game much faster if everything was automated (you just put how many players you have that are part of the same team, what skill level you are and whether you have a server. The system matches you with a team within your skill level and will favor teams with similar number of team members. It could also work out a suggested div to search based on the players’ divisions.

Also trath its really cool that you are doing this, have a panda video to spur your efforts: http://www.youtube.com/watch?v=ayCRfkGA0qE

Currently I am not planning to give weight to a certain class, I am going to do it the way TheSucker suggested, it seems like the best way to me. (Which in a certain way is like weighing the demo and medic more, and in another, not).

Obviously, nothing is final. I just know what I am aiming for right now. If after a month or two of the bot being up, it is clear that something in the dividing of teams or changing of ELO upon win/loss needs to be changed – it will.

I only sort of understood this mix bot you are talking about. I’m going to finish this bot and get it up(hopefully on Pixel), then not touch anything related to that shit for at least 2 months. I miss my life and actually playing TF2 :( I just feel like there is a little “hype” for this bot, so I don’t want to slow down now, only get it up in 3 weeks and then everybody will forget about pug-ing.

*UPDATE: I am 99% sure that the balancing/ELO system is complete, but would require 12 humans to try it out and confirm.

Skeej:
While I do generally agree that humans are always better than a piece of metal, I think it is not the case here.
If a pickup chan like that actually got popular, the new guys / div 4-6 / fat kids will always stay fat.
Also, I think its safe to assume that a popular pickup chan would have too many players for one person to recognize and actually know their skill level.
Making captains pick teams is also something that requires more time/waiting.

Your argument that a pickup chan should not require auth is not an argument. People need to identify themselves, and auth / Q allows it so. Who doesn’t have a Q account? and even if you don’t, it takes less than 2 minutes to sign up and add the auth to Remote so that you auto-auth upon connection.

TF2 lobby requires you login. ETF2L requires you login. Should we make a new league that requires no login? I’m pretty sure 1/2 the players will be “Schocky”.

You should give this balancing bot a chance, I think you will be surprised.