TSL4 Changes Discussion

[qan]

Looking at some changes for the next TSL season.

Chief complaints seem to be:

1. Ends up being mostly reccing (especially in j/d). The vets rush to pick up the easy kills first for a chance at a top score, and then finish one another off at the end.
2. Ratings are sometimes inaccurate.
3. Matches can play unevenly, even if skill balance between the two teams is fairly good, as it throws a lot of different players in the mix.
4. Finals don't feel meaningful without recognition of who has won or lost.

(Have more complaints? Post!)

So far, some ideas to address some of these issues:

1. Value all kills based on the killed player's rating (in j/d). It doesn't make sense to value 15 kills on top players the same as 15 kills on guys who started playing yesterday. Some kills are much harder to make, so giving more value for those kills would align value more accurately with actual skill. How much reward is given for a kill should match the difficulty needed to get it as closely as possible. That's the definition of accurate value. This would require some serious work to get the curve right, probably starting with the average rating in a league being worth 1.0. Alternate approach: Downscaling of value for kills based on your rating vs player killed. If you're far above another player in skill, the kill won't count for as much as someone who is roughly around your rating. This has the problem of boosting lower-rated players.
2. Looking at implementing floating ratings with a simple system based on a combination of a strong performance and winning the round. The one problem here is with unrated players. Do we just set them to 40, and then modify their rating based on that? (By doing so, it makes it difficult to see which players were actually unrated, and which have been set manually to 40 by ops.)
3. If we implemented solution #1, this could allow two tiers of games, when 4+ teams can be created. A high-skill game and a lower-skill game. I'd have to see how to implement this in the brute force algorithm, though. Might be tricky. With any luck, a change in kill valuation would also affect the uneven play of matches, and it wouldn't be as much of a problem anyhow.
4. We'll be awarding champion medals on the TSL maps, not just on player profiles. Would like to find a way to make finals feel more meaningful, too.

(Have more, or better, solutions? Post!)

Current plans point toward a sort of hybrid with TSL/TWD, focusing on hosting a small bit of TSL, and then handing it off to staff-organized TWD matchups, where squads can show and get players to borrow for games (TWDBot's !borrowlist and !borrowme), a bit like TSL matchmaking works, but using existing TWD tools and squads. Some of the borrowed players should be recruited to the squads as well, with the new in-game tools making this much easier for a player new to TWD to get started.

[Tiny]

If I'm not mistaken, the elim bot did what you were talking about. Killing people with higher rating got you more points, lower rating, less points..etc...Look into that.

and yes, split the games based on rating......16 plyers...take top 8 and split into two even teams....bottom into same....ez fix

[Tiny]

I say easy fix, prolly means a bunch of coding 8)

[Tiny]

let rating change more than a +/- 1/2 variance.....make it open ended...

[qan]

If I'm not mistaken, the elim bot did what you were talking about. Killing people with higher rating got you more points, lower rating, less points..etc...Look into that.

Here's what I came up with for #1. Think it works decently. Average rating for d/j (after ignoring unrateds) is almost 60 exactly. So if we use a multiplier of 0.017 on the rating of the player you killed, it works out. Your average, everyday kill should be worth 1 kill after rating adjustment. So 60 x 0.017 = roughly 1. Lowest value is around 0.68 for an unrated player, and 1.7 for a 100-rated. Seems a bit high, but then, sure, if you make a kill on Ease vs a kill on joepubber42, it doesn't seem so unreasonable that it should be valued more than twice as much (honestly, probably far more). We could look at a cap, though, maybe max 1.5 value or so, as it might lead to excessive hunting of high-rated players. Essentially the reverse of what we have now. That's far, far better than newbies getting hunted, as at least hardcore vets can defend themselves. They might even enjoy the challenge, and be able to use it to bait people into making errors they could then take advantage of. Sounds far more interesting than everyone rushing to eliminate the weaker players, and that being the most substantial part of the game -- whoever is lucky enough to make a bunch of easy kills at the start gets the best score.

and yes, split the games based on rating......16 plyers...take top 8 and split into two even teams....bottom into same....ez fix

Actually don't think this would be that tricky. It'll result in teams that aren't as balanced. But still would create much more interesting games.

let rating change more than a +/- 1/2 variance.....make it open ended...

Problem is, the more you allow rating to change based on a single game, the tighter the formula has to be. And let's be honest, not even Ogron was all that good at this kind of thing. Any kind of complex formula needs a lot of testing to prevent edge cases from causing problems, places where the intention of the formula is defeated by extremes. In TW, all the testing is live because we have no QA department. Hell, hardly have a dev department. So if it goes wrong, it could potentially screw up the entire season.

But, if there's a very simple option that could allow for greater ratings changes, definitely willing to try it. Just haven't seen that yet.