TWDT Star Caps: Discussion

[Rab]

In that case, see this post: https://forums.trenchwars.com/forum/...ap#post1331030
This was my attempt to create a formula to calculate a star cap.
The general idea is:
- if players were distributed between teams evenly based on rating
- then each team decided to field it's top 5/8 players, that produces a star number (which represents no limit if an ideal draft occurs)
- optional: then reduce that number by some amount to allow some players to play at the expense of other players
I think a formula is better than guessing.
The formula requires you to have a complete list of signups with ratings first.
I think you should work on improving the formula so it can be re-used for future seasons with different signups/ratings.

[Poseidon]

if theres going to be star cap my 2 cents:

- Basing goes to 5-10*
- Re-think star cap based on 5-10 system
- Allow basers 1-extra ship rating if they request

[ogron]

The whole point of 6*-10* is that players are actually 1*-10*, but you can improve dramatically over a short period of time from say a 4* to a 6*, or you're a relatively unknown player from another zone and misrated and you're actually good.

Putting everyone in that 6* basket mitigates the damage. Kassius was misrated in base and the best 6* in TWDT-B, but it was manageable in terms of playing Terrorist.

If we had a 1*-*10 system and he got rated 3*, he would be so unbelievably broken it would be disastrous.

Setting the lowest level to 6* is smart and there for a reason. Every season there's a few players who are one star off and people complain, but it's super manageable. Teams win TWL games with multiple fewer stars on their team. There's always going to be good teams with economical star values, but the point is that it's still winnable even if they're favored. It's always playable. A big part of that is not giving out 1, 2, 3, 4, and 5 stars. They cut out 5* based on experience. It used to be 5* to 10*, and based on years of play there was discussion and agreement on removing the 5* and setting the mark to 6*. There was some inflation, but I think the overall effect was positive.

[Poseidon]

oki well if uve tried it and it didnt work for some reason then nm. feels like more gradient the better. for instance, if last szn was re-rated on 5-10*, zidane is prob not as likely to get 6 understarred spiders, or at least the effect wouldnt be felt as strongly.

regardless, i do hope ppl seriously consider the other proposal which would allow people a 2nd ship rating... it would add a nice bit of depth to drafting and the league itself; a lot of players would be grateful for it, and captains as well as it could greatly increase the flexibility they have with their lineup during the season

[qan]

One reason that star values are flawed is purely mathematical. The relations between star values are not accurately quantified. This becomes increasingly obvious as you assign lower star values, and is why using values <6 results in imbalance, especially when using an overall star limit.

We need a better way to model what a star actually means.

The difference between a 5 and 6 star should be such that (with accurate ratings):
- If you decided to play a 5 star instead of your 6 star and a 10 star instead of your 9 star (trading out one star on either side), the overall skill level of the team should remain relatively unchanged.

This means star should be mapped to another quantified skill level which is defined by a function. I'm messing with a graphing calculator right now, can't quite find a great function to mirror the desired behavior off the top of my head, but it's worth messing around with if anyone has some free time and remembers some of their high school mafs.

[Turban]

in my opinion we should either introduce half ratings to player ratings to properly differentiate players from one another, or alternatively switch back to old rating system where players were rated on their experience and skill level while also limiting number of players that can be in that slot. it was something like this:

10-star would be top 3 (or top 5) in a league.
9-star would be top 10 in a league.
8-star would be potentially impactful TWL/TWDT starter.
7-star would be someone who consistently starts in TWL, but you can't expect big number games more often than maybe 5% of the time.
6-star would be TWD player with very limited experience in top competition or players from other zones with no prior experience.
5-star would be absolute noobs from trench wars.

however, the first option seems better. i feel it would be much better if we could rate players to be 6.5, 7.5, 8.5, or 9.5. team star cap would be rounded up (39.5 would be 40).

[Poseidon]

.5 option feels nice!

qan or someone - would it be simple enough to have TWO ratings per baser? one is the Primary Rating or their highest rated ship; the second a ship of their choice, only per request. Primary rating = rating of all ships aside from the secondary choice.

[WillBy]

.5 option feels nice!

qan or someone - would it be simple enough to have TWO ratings per baser? one is the Primary Rating or their highest rated ship; the second a ship of their choice, only per request. Primary rating = rating of all ships aside from the secondary choice.

Qan can probably elaborate, but the current bot star counting system wouldnt work with multiple ratings. That feature could he disabled with staff manually counting instead.

[Rab]

Qan can probably elaborate, but the current bot star counting system wouldnt work with multiple ratings. That feature could he disabled with staff manually counting instead.

https://forums.trenchwars.com/forum/...15#post1330715

[Nipple Nibbler]

.5 option feels nice!

qan or someone - would it be simple enough to have TWO ratings per baser? one is the Primary Rating or their highest rated ship; the second a ship of their choice, only per request. Primary rating = rating of all ships aside from the secondary choice.

Some of us play all 3 ships, if we are going to separate the ratings you may as well go to all 3 not just 2.

Also +1 for .5 ratings, we have needed these for awhile there are so many of us that are half way between numbers.

[qan]

Yep, as Rab quoted -- that analysis still stands. Subs and switches get tricky if forcing star caps. We're always doing league botwork last minute as we change all the rules, which is absolutely insane, but also kind of the norm due to it all being volunteer-run.

[Gkyde1234]

1/2 ratings can be abused by raters. Last time we tried that I got rated 7.5 an could only play the 8 star slot. Keep that in mind if you get bumped up .5 this year

[qan]

I talked to an old friend who's got his math PhD and he recommended I play around with a formula along the lines of this to arrive at real star value. A value that accurately represents a player's skill that is extracted from their star value. If you don't understand why this is important, please ask.

Here's what he suggested I start with: C1 * e^{(C2 * stars)}

e is Euler's number, 2.71something. C1 and C2 are small numbers in this case (<1 seems to yield decent results). For an example I'm using .8e^0.2526x

Notice how you get a smaller difference between star ratings the lower the rating goes. It might be useful to add an offset to this value in order for it to work in practice. For instance, +10.

Stars	Adjusted value (approximate)
1	1
2	1.3
3	1.7
4	2.1
5	2.8
6	3.6
7	4.6
8	6
9	7.8
10	10

NB: I don't think these are the right numbers (for example, look at 7 vs 10) but this is what should be done to arrive at an adjusted star value.

To play with it, tweak C1 and C2. (Which in this link are set at .8 and 0.2526) Mouse over points on the graph to see exact values. Anyone can do this, it just requires monkeying around. Our current linear function where x=y is garbage, though, and needs improvement. This is evidenced by the fact that half of the numbers are completely unusable. I don't hold out a lot of hope that the community would opt to change even with the right formula, because let's be honest, we're a pack of very stubborn mules, but I'd at least like to make clear what the issue is so that we understand what we're dealing with.

Pinging the numbers crew: Turban Claushouse Rab Nipple Nibbler @anyone else I forgot

https://www.desmos.com/calculator/fxc1xojlfi

[Rab]

If I was going to change the players rating to anything I'd want it to be something automated that completely removes the need to have this discussion ever again. The fact of the situation is that none of our ongoing stats actually represent us because it's either twd that is too casual or twl that is more of a reflection on your team than on you.

So I think using our judgement is still the best way we have. That doesn't mean it can't be improved though. We have a really good stats spreadsheet from TWL so we should use that as a basis for ratings adjustments.

[Poseidon]

Rab Good analysis -- that's all true. WingZero was the man who took on the job of reworking how TWDT is handled by the bot/in the db. Changing to have three different ratings is fairly trivial (assuming specials use Spid rating), but the issues with subs and switches aren't.

Validating manually might be the easiest solution. Have the bot print out current star totals based on the ship in play, with arena message and PM if over the limit indicating that keeping the line as-is would result in an invalid line.

quote from qan in thread cited. basically what im reading in that post is that it would be hard to code multi-ship ratings and have it all run smoothly. theres an easy solution however:

if someone wants a second ship rating, they must make a new name. they would play the ship with that name, and it would be a separate star. the question then is: would it be difficult to program the bot to say "this player may only enter game in ship X and none other". If this is too much trouble... simply make the team forfeit the game if they try to cheat or "accidentally" put a 10*spid player in spider instead of the 8* shark their name was registered to. really dont think we'd have any problems.


Nipple, the reason for allowing only one extra rating, and only when requested is multipart:
- it helps the whole thing from becoming too convoluted
- makes it easier on raters
- makes it easier on captains doing line/drafting
- the limiting factor serves as a unique strategic element in ways
- considering 2 would be easiest to implement and integrate, its sensible to first see how it goes with 2 and later explore the idea of 3+ in future seasons