TWDT 2 - a proposition

[Exalt]

Didn't we already go over this lol. My win lose rate on the overall stat page crushes you and I just crushed u again here tonight. It is more fair to every season have a battle to earn spots in TWDT.. and use W:L stats as part of overall stat placement if you want. Noob

Did you really just post a random dd and think that somehow shows something? Your stats are meaningless, because you lose almost 75% of every game you play in, the 2nd worst on your squad behind Lens who is like a 4* wb. You are the type of player that gets stats by making their team worse.

It's not like this is a big secret though. Everyone in the game already knows what I'm saying except you.

[qan]

As an aside, in proper Elo ranking (also as it has been adapted to many video games), nothing matters other than a win or a loss.

TWD players could have the same style rating as TWD squads, really. It's a very simple but effective formula. You can get 21-10 every round, but if you win 49% of the time over a very long range of time, aren't you technically a very slightly worse player than average?

[beam]

not sure where those #s are coming from but according to twdd stats jessup has 33 wins 32 losses and exalt has 68 wins 92 losses

[beam]

As an aside, in proper Elo ranking (also as it has been adapted to many video games), nothing matters other than a win or a loss.

TWD players could have the same style rating as TWD squads, really. It's a very simple but effective formula. You can get 21-10 every round, but if you win 49% of the time over a very long range of time, aren't you technically a very slightly worse player than average?

in an elo system you are assuming that you will face players at your elo which is simply not the case in this game as a squad like paladen will lose to literally anybody else if I faced paladen and rocket every game i could easily be 10000-0 in twdd

[Exalt]

not sure where those #s are coming from but according to twdd stats jessup has 33 wins 32 losses and exalt has 68 wins 92 losses

[Exalt]

in an elo system you are assuming that you will face players at your elo which is simply not the case in this game as a squad like paladen will lose to literally anybody else if I faced paladen and rocket every game i could easily be 10000-0 in twdd

If you played on Rocket or Paladen you'd easily be 0-10000 as well. That's why you have to look at only the squad's win% and stats.

The difference would be instead of you having a 34% winrate while obviously better teammates who have played similiar amounts of games have much higher %, you playing on Paladen or something would likely show you having a higher win% than the other players, simply by virtue of them losing far more when you aren't in than when you are.

[qan]

Would have to look into the details, but it may simply be a matter of adding the two teams' ratings together, and comparing those with the formula, with perhaps some factor influenced by the number of players (or similar variable). Whatever it is, it's a system used in many places where team games are played. Details are likely out there, if anyone would care to research. I did some very very cursory stuff.

But wins over the long term (with a large enough data set) is really the only true gold standard measure of skill in a team game. Whatever skill you have, it will be manifested over the long term if you have a higher percentage of wins in your team over teams of similarly-skilled players. If they are not as skilled, of course your rating goes up only a slight bit or not at all, directly in relation to how big a difference there is between ratings.

[Exalt]

Would have to look into the details, but it may simply be a matter of adding the two teams' ratings together, and comparing those with the formula, with perhaps some factor influenced by the number of players (or similar variable). Whatever it is, it's a system used in many places where team games are played. Details are likely out there, if anyone would care to research. I did some very very cursory stuff.

But wins over the long term (with a large enough data set) is really the only true gold standard measure of skill in a team game. Whatever skill you have, it will be manifested over the long term if you have a higher percentage of wins in your team over teams of similarly-skilled players. If they are not as skilled, of course your rating goes up only a slight bit or not at all, directly in relation to how big a difference there is between ratings.

Hey Qan, quick question btw, what is the star rating on the twd site, and why is everyone's so low?

Edit:

Here's Rocket's stats page for TWDD. How does the star rating thing in the 6th column work? Seems wonky to me.

[Jessup]

not sure where those #s are coming from but according to twdd stats jessup has 33 wins 32 losses and exalt has 68 wins 92 losses

Thx beam . Exalt is making stuff up.

qan Qan while I agree that W:L ratios are important stats to look at in determining player values it is not the only stat that matters. Kill/ death ratios and kills per round averages and time played and KO's all play a huge part in a players value too. Only looking at win/loss ratios can be VERY misleading if the line you play with is inferior to the line you face. Good players who play on lower tier squads will suffer if their line consists of lower tier players. It doesn't make sense if say Lockdown played on a last placed squad that consistently used a majority of low tier players to think his win loss ratio would be in anyway reflective of his personal skills and attributes towards his team. Naturally if you have 1 or 2 high tier players and 3 low tier players the results of win/loss ratios would be greatly impacted and vice versa for those who consistantly play on high tier squads who only verse low tier squads. The rank of your squad and the type members it has will drastically effect these numbers and you can not put any real credence focusing only on W:L ratios. Now if there was a wiped clean slate and ELO was started in an evenly matched tourny with as even as possible lines fighting each other then it could hold more value but like elim there should be many stats working together to place a truer value on each player.

I also want to apologize to MHZ that their thread got derailed some too. All I really wanted to say here is didn't want to vote on this idea because I offered an idea of my own how to deal with TWDT by getting rid of star values completely and instead having a pre twdt competition where the people who do the best get to play. IMO the star system is broken.

[Jones]

Can you guys stop ruining this thread? Im trying to figure out how i can battle for aml medals here

[Exalt]

Can you guys stop ruining this thread? Im trying to figure out how i can battle for aml medals here

1. Be rated under 8*
2. win
3. profit

[qan]

Hey Qan, quick question btw, what is the star rating on the twd site, and why is everyone's so low?

The intent with the original design was to input current DT star values into a player's rating at the start of a season to show expectation of players' performances, as we don't have a large enough number of games played in a season to allow floating ratings to do their work properly. When this was implemented, we weren't storing star values in the database, so it was an extremely tedious by-hand process and was discontinued. (This goes some way to explaining why ratings don't work so well currently.) BIET or others might be interested in pulling star values now that we do store them.

The star values then float based on performance. So whereas when doing manual ratings we never rate anyone below 6 stars to keep from hurting feelings, the rating system has no such qualms. I would not say the floating algorithm is all that robust, though. (For one, I think it probably floats too much, but I'd need to review to be sure.) Unfortunately, having a rating, star rating and adjusted rating is confusing. Rather than including star rating for expected performance, it would be better to simply use a zero-sum system and have K values (how much a rating moves based on performance) stay high early in the season, and lower later in the season. A bit like how most games use qualification matches. These are just matches with very high K values used to quickly but very roughly gauge skill.


I did a slightly deeper dive on team Elo. In Chess, where Elo ratings come from, you have only Win/Lose/Draw. With a game with more complex win conditions, such as a score, time held, etc., you can still check for margin of win. The problem is figuring out how much that margin of win accurately represents skill. So how much more is a 15:00-5:00 in base worth more than a 15:00-14:30, for instance. Getting that right determines how accurate Elo is at representing actual skill. There are other problems with combining player ratings from a team and applying the team's + or - of rating to their rating, but aside from being a bit complex it's a problem that many have thought about and implemented solutions to. Basically, such ratings are generally more predictive of actual skill over time vs simple W/L ratings or complicated systems designed to measure skill through careful formulas of individual stats (which is what we currently use). That last method is potentially the most accurate, but also the most difficult to get right, and it can create some very serious issues when it allows rating to consistently increase even when a player is losing <50% over time vs similarly-skilled teams. (This opens up the possibility for reccing to the detriment of your team.) If a player does well but their team still loses vs a similarly-skilled team, that player's rating should still go down, but arguably by a smaller amount. That would be another point to get right.

[maketso]

[May 02 21:05] Commodo: you just gonna vote down my forum poll and ghost eh? fk u bud.

I wouldn't do any big changes to the league that seems to be working just fine