Proper ranks

[dansan]

[..] big team [..]

Actuall it's just "team", regardless of 2v2 or 16v16.
I was thinking of making 2v2-4v4 and >=5v5 separate, but that would also lower the amount of data for evaluation, resulting in a slower conversion on a proper rank. And the question is also, if it would actually make such a big difference...
Maybe a solution could be to use a common value (like now) until for each category (<5v5 and >=5v5) there were 15 matches, then separate...?

[gajop]

why not just have a single elo ladder for teams > 2 or 3 people, and just have a smaller elo impact per person the bigger the games get ?

[dansan]

why not just have a single elo ladder for teams > 2 or 3 people, and just have a smaller elo impact per person the bigger the games get ?

It's already like that.
(Elo is used only for 1v1, TrueSkill is used for Team and [Team]FFA)

[Jazcash]

What I would like to see is another ranking system that's generated from a cool algorithm that takes lots of battle data into account such as damage done, damage taken, micro efficiency and all that jazz.

Ofc these factors aren't the be all and end all of a player's skill and a sometimes they'll be silly because they'll all go to a player who eco'd all game and just won with T3 spam or w/e, but I think they'd be nice to see, nonetheless.

If there was some way to extract all that kinda info from replays and made accessible as an API or something I'd love to have a tinker and see what I could come up with. Spring Replays API plzplzplzplz

[smoth]

Jaz from you I would expect better "micro efficiency" is too abigous/broad a concept to really be meaningful

[Jazcash]

Jaz from you I would expect better "micro efficiency" is too abigous/broad a concept to really be meaningful

?

I was referring to the type of micro-efficiency in the post-game autohost reports:

[2:59:47 PM] * [TERA]Host1 * Damage award: Crossaro (total damage: 1008K.)
[2:59:47 PM] * [TERA]Host1 * Eco award: [AOE]ShocKwave (resources produced: 49741K.)
[2:59:47 PM] * [TERA]Host1 * Micro award: Crossaro (damage efficiency: 257%) [ O

Hmm I guess damage-efficiency is the correct term, but w/e

[dansan]

What I would like to see is another ranking system that's generated from a cool algorithm that takes lots of battle data into account such as damage done, damage taken, micro efficiency and all that jazz.

IMO it'd be nice and fun to have those ranks, but I don't think they are necessary for balancing.

Alba wants to do just that for some time now. I'd love to help with it, but my thesis is taking much longer than anticipated

If there was some way to extract all that kinda info from replays and made accessible as an API or something I'd love to have a tinker and see what I could come up with. Spring Replays API plzplzplzplz

The problem is, that the data must be collected at runtime. See http://springrts.com/phpbb/viewtopic.php?f=16&t=29187

Also see knorkes demonaut (http://springrts.com/phpbb/viewtopic.ph ... 71&start=0). Produced battle reports like this: http://knorke.net23.net/battlereports/E ... g.txt.html
It's a widget that collects data during the game and a c++ program that creates the statistics and the resulting webpage afterwards.

[smoth]

[2:59:47 PM] * [TERA]Host1 * Damage award: Crossaro (total damage: 1008K.)
[2:59:47 PM] * [TERA]Host1 * Eco award: [AOE]ShocKwave (resources produced: 49741K.)
[2:59:47 PM] * [TERA]Host1 * Micro award: Crossaro (damage efficiency: 257%) [ O

Hmm I guess damage-efficiency is the correct term, but w/e

What the hell is damage efficiency? Damage/cost etc? so the guy didn't build much econ etc? What are the actual metrics you want to see

[Jazcash]

What the hell is damage efficiency?

I'm guessing it's just similar to unit experience points? Perhaps even based on that, I'm not sure. Basically, you're not gonna get any points if you make a bunch of units and do hardly any damage with them.

[smoth]

so it should be damage/cost.

damage efficiency would imply units were used in ways that best utilized their armor type advantage.

Secondly, when my next project comes out, I am dumping the spring code for unit experience and tracking it differently. So using any of that rubbish would be unusable for any project using my work. TOO MUCH dependency on using the internal built in stuff on a metric that is supposed to apply to all projects.

Consider how the request MAY or MAY NOT play out outside of *A please.

[knorke]

The problem is, that the data must be collected at runtime. See viewtopic.php?f=16&t=29187

hm but chatlog can be extracted. (there is a tool for that on springfiles)
So at end of game you could just dump all the collected data into chat, which is what zk/spring does I think?
(open a replay in texteditor and there is things like SPRINGIE:stats,dmg,armbrawl,armwin,734.97583007813)

[Silentwings]

We should not rank people on anything other than win/loss because we shouldn't be telling people what is/isn't important about how they play the game.

Yes this make convergence slightly slower, but (i) trueskill converges pretty fast anyway (ii) we can still have chranks (iii) we've survived for years on a far worse ranking system and it hasn't been all that bad. More important is to get trueskill built into spads and then decide if we still have problems to solve.

[smoth]

We should not rank people on anything other than win/loss because we shouldn't be telling people what is/isn't important about how they play the game.

Yes this make convergence slightly slower, but (i) trueskill converges pretty fast anyway (ii) we can still have chranks (iii) we've survived for years on a far worse ranking system and it hasn't been all that bad. More important is to get trueskill built into spads and then decide if we still have problems to solve.

I agree but I am still interested in the stats jaz was asking about because I can try and come up with a custom end game screen with it :).

[Silentwings]

...stats ... custom end game screen with it :).

I like

[smoth]

I already started code to collect data, I just moved to other bigger tasks :)

[Jazcash]

If you replace the post-game graphs with a smexy Lua alternative I'd blow a load

[smoth]

Right now what I want to do first is a LOL style end of game penis grid.

[dansan]

The problem is, that the data must be collected at runtime. See viewtopic.php?f=16&t=29187

hm but chatlog can be extracted. (there is a tool for that on springfiles)
So at end of game you could just dump all the collected data into chat, which is what zk/spring does I think?
(open a replay in texteditor and there is things like SPRINGIE:stats,dmg,armbrawl,armwin,734.97583007813)

Uh - that is ingenious! Keeps the replay and the stats together in one file.

More important is to get trueskill built into spads and then decide if we still have problems to solve.

+1
Call to all devs that know a little perl: please help vbs in that - together I'm sure it can be done fast!

If I understood him right, he's doing a plugin-system, so that different (multiple?) backends could be used. If we get this, then everyone can just code away and easily test her most crazy ranking-algo.

[very_bad_soldier]

Call to all devs that know a little perl: please help vbs in that - together I'm sure it can be done fast!

I kinda stopped working on it due to frustration about TERA-8v8DSD and smurfing in general. Seems impossible for me to find an enjoyable game nowadays.

[muckl]

1. real trusted user account system to prevent smurfing
then
2. do your auto ranking system

otherwise its fighting windmills

i think smurfing is way bigger problem than not having autoranking
cause at the time smurfing is useless or not usable, then manual balancing is good too before we get some nice autobalancer