It was an absolute blast having you Dean. I hope I'll see you again! I think my favorite moment was immediately after the Brawl finals, you turn to Aero and go, "I hope that was as fun for you as it was for me." I died.
SpaceJam said: oh my god stop writing such long messasges.
Can it become a rule that everything has to be communicated in emojis?
I was trying to be thorough, but apparently that's "offensive" lol. But fair enough, point taken. Shouldn't be an essay.
MrL, no one is taking "jabs," but you continue to again not hold yourself to your own standard. You expect me to answer your questions when you ignore mine, and fall back to generalizations (FYI, Elo is not "transferred" or "shared" between games). Even this "put words in your mouth" bit is silly. I quoted you, but you actually do put words in my mouth. This is why examples and hypothetical situations are important to be clear. For example, when did I ever say every permutation of teams would be listed? I can and have quoted you putting these words in MY mouth, but I don't see when I did that to you.
Do unto others. I'm not taking jabs. A disagreement is not an insult; neither is a misunderstanding. I'm not even against your idea of teams, but was saying, "this is great, we should rank teams! Let's do it, and while that's going, find a way to cover all team matches!" If you didn't get this from the get-go, then it's a misunderstanding, not a jab.
MrL said:You're raising issues that don't really exist. The hypothetical situations you raise are pointless.
You say that, but you didn't address anything I said at all. :-(
MrL said: In your situation... there are now 20 teams that Jaice is a part of on the list of teams (just because of one person, let alone the 50 others who may switch between partners; lots of clutter).
That's not true at all, and even if it was, do you see the irony here? This is fundamentally the problem with your approach, and hurts your "registered teams" method. Your solution is, "well, this is convoluted, so we ignore it." That's not a solution; it's a "la la la I can't hear you I will ignore this" approach.
You know how we don't list players on the Elo list if they haven't played for the past six months? We could easily do the same here, if not just have a set minimum of matches. Hell, the beauty of this player-centric way is that it doesn't need to be listed at all, but all those results would be acknowledged on a player's singular page. Your strawman above is akin to saying "we have many people who never attended a tournament, so why are we listing them at all?!" The answer is, we don't, so why are you saying we do?
MrL said: But there's also no way to determine which time is better, other than a theoretical algorithm.
I've read this a few times, and I don't get it. You say you don't have a solution, so you are complaining about a potential solution that we can continue to tweak and improve. No one is saying we do any sort of overfitting or underfitting, so I'm not really sure what you're getting at. An algorithm is indeed the solution, so there is a way. O_o
MrL said: It may say that one team is better than another, but only in theory; the teams would need to face each other to determine that, which can't happen. Then this process would have to happen with every person.
You realize that is what Elo is all about, right? How does the same exact statement not apply to your approach? Elo is LITERALLY a predictive score. It is a guess about, when two entities face each other, who will win. When it's right, it reaffirms the distance between the entities; when it's wrong, it reduces the difference. Elo is "in theory," as is your registered team Elo.
Why else are we using Elo? Why are you using Elo in your registered teams approach if you are arguing that the theory of Elo is not acceptable by your standards? Elo was in fact designed to predict hypotheticals, even if they never ever happen, like two chess masters who were separated by generations.
MrL said:It not only would just look ridiculous to have one person in every (or even multiple) team(s) as 'the best teams' but also impossible to determine which is the best amongst those as they cannot battle eachother.
Yes, there would be n * (n-1) potential teams, but your registered team ignores the vast majority of them. We don't have to list them all (see above), nor do we calculate their scores either (that's the beauty of a player-based approach!).
Your method: I don't see how ignoring these teams represents anything. Do you just want a handful of teams that don't represent anything but a selective group that has no predictive value - or any meaning for that matter? If you're saying one person can only register with one other person, that's a "la la la ignore" solution.
MrL said:No data will ever be 100% accurate, players and results are constantly changing, and at any given time someone's elo may not represent their skill.
That's what Elo is all about. That's why it changes. How does your approach "100% accurate"-ly predict changes?
MrL said:With set teams, there are constant variables that are pit against each other over a period of time.
Great! Can you specify what these are? Bear in mind you just said above that "results are constantly changing" and Elo "may not represent their skill" but somehow your registered team is immune.
MrL said:Within this set of teams, a ranking order would be able to be established (and one person won't be in multiple teams that will be unable to face each other).
From this, a more accurate PR can be established.
How can any PR be established? I've explained this before, and you did not address it. Your Elo approach wouldn't represent anything after more and more results are acquired. You're free to use the example I gave, of Bob the Builder and Timmy the Tester scenario. This is the value of the hypothetical situations you quickly dismissed without explaining why you dismissed them. It explains why your approach will fail over time. How, in those situations - which would eventually happen as the number of doubles matches increases - will your solution continue to be "100% accurate," "not in theory but in practice" best team ranking order?
MrL said: Using that algorithm and making a new team and scoring them based on who is teaming, that will only result in the top 10 being 5 teams with jaice and 5 teams with jezmo.
I think you're making assumptions on Jaice and Jezmo's team variables, as well as other players they would team with. Again, while my equation is just an example and wholly incomplete, it still doesn't suggest this in the slightest. Feel free to punch some numbers in. Talking in the abstract may be the reason for the confusion.
MrL said:However, I'd rather a few one-off teams slip through the cracks than have an entire list of teams be made up of one person holding the top several ranks with [insert random players]. Again, even with that list which would look ridiculous, there's no way of determining which is better among them.
You're not getting this if you're thinking this way. At a certain point, the "registered team" Elo doesn't represent anything but a biased sampling of people. You ignored it the first time, so I'll say it again: the more results you have, the LESS accurate your approach will be (see Bob and Timmy). You haven't demonstrated a solution to this. Instead, you ignored it and just said there's no way to figure out with my method (although there is). What part of Bob and Timmy is not making sense? And how would Jezmo plus random player even be high? Have you actually plugged in Jezmo with a random 1000 Elo player in my equation? Again, my equation may not be perfect, but even it doesn't rank them high.
MrL said:It's just not viable to have a system that tries to rank every combination of players, as that way the teams that play against each other will be different each time and impossible to accurately show improve because of the number of variables (hence why we don't already have it), let alone one person can't be on two teams at once to determine which of the two (or more) they are a part of is better.
You can do it all hypothetically with the algorithm as said but that's not reflective of actual doubles results between the various teams. I just don't see your way as a viable or accurate way to determine the best teams.
Okay, no more strawmen after this, okay? I never said we rank "every combination of players," and thinking that demonstrates a complete failure in understanding my player-centric approach. You need to explain how you "solve" all the "issues" you've raised. It's really not acceptable to say, "I'm okay with all these problems to the point it becomes meaningless to have team Elo." Why would qldsmash implement it then?
MrL said: Prior performance with other people, not the same person.
I don't see how it omits one particular partner. It's not like playing with the same player multiple times will be ignored. If you mean every person including the same person consistently, then yes, of course. That's why it's more malleable than the "registered team" approach.
MrL said:I think the doubles elo could just as easily and more simply be calculated based on how the teams interact with eachother, with the players remaining the same, rather than individually calculating each team's capability by calculating each player's elo, characters and prior performance with different people as separate factors.
Yes, registered teams could suffice. However, it's a quick fix and will be biased as hell as more doubles events occur. You are literally selecting for people who pair up frequently, even though they may be the worst doubles teams. Imagine Jaice cycling through all the top 20 players in 20 tournaments, but Bob the Builder team is the highest ranked because they were the only ones registered. Or the last 15 tournaments, Bob always placed last. Yet, since the Builders didn't play any registered team for the past 15 tournaments, their Elo stays intact with a win rate of 20%. Meanwhile Timmy the Tester's group won the last 10 tournaments but only beat unregistered teams and is calculated as worse than Bob's. As you can see, it's a contrived approach that only works as a short-term solution.
I don't see why we can't start with it and leave it around. But long term, we have to come up with something else, since it will strongly result in selection bias over time. Basically, it begs the question, if it no longer even reflects Elo, why even have it?
MrL said:Just how singles elo isn't calculated depending on what character we use and prior tournament placings, but rather who we face each time and whether we win and lose against them each time.
Just because we do something some way doesn't mean it is correct. Guns are legal in the USA, so Australia must be wrong? I do think character matchups should influence points. If you win in a tougher matchup, you should get more Elo; in an advantaged matchup, less. However, 1v1 matchups are more contentious than "good doubles characters." But that's a tangential conversation, so we'll leave it at that.
MrL said:The individual skill of each player plays a role in that yes, if you have better players teaming then yes in all likelihood they will do better, relying on their performance as individuals to figure out team performance won't nevessarily work.
I don't understand what you're saying. So it does influence or does not influence team performance?
MrL said:Would it not be better observe how the team performs, as a team; rather than calculate their skill using their individual performances as the key factor.
How is individual performances the key factor? Can you show me what you mean? Do you just mean "the biggest number?"
MrL said:To me it seems like its overcomplicating it when you can just start on a clean slate and see how people perform from there.
Skill doesn't "reset," but okay, let's pretend it does since it doesn't matter in this conversation.
MrL said:When wii u elo was first calculated, did everyone start at 1,000 again or were Wii U elo's originally calculated based on the limited 3ds elo data?
Strawman fallacy: the logic here is akin to saying people's Melee data should be used in Smash 4. I don't recall making either of these statements. 3DS is a different system, and wireless connection has input lag because of variable speeds. I guess you're saying that diffusing a bomb with a timer and without a timer requires exactly the same skill sets with no new variables, but I'd disagree. Besides, it doesn't really add to the conversation.
What? The calculation ***is*** based on team performance. I even gave an example of just that.
Of course the Elo would be based on individual performance, but the other variables would not be. And it's a little incredulous to think individual performance does not influence team performance whatsoever. If the extremely large dataset from M$ is any indication, the change in individuals within a team is not an impossible thing to solve.
The lack of constancy in teams is the reason I recommend a player-based calculation. All the values are based on the player (and characters used), not on a single combination since teams change. This is very much like Microsoft's approach in the sense that their algorithm could predict outcomes really well for people who have never played together before.
SpaceJam said: Nekminnit you'll be saying Bowser isn't good in dubs...
Damn straight. But yeah, for example, very crudely, I'll give a hypothetical, sans data equation:
Team Elo = Total Elo * a^2bc / 2
a = How good a player is in a team (so a1, a2, but just doing a^2 to be obvious) = 0.0 - 2.0 (realistically, 0.8-1.2) b = How good their characters work together = 0.0 - 2.0 (same as above) c = Variance in skill between players = (1 - [P1 - P2] / Total Elo)
The 2 is just a correction factor for "teams."
Now let's say it's two people with Elo of 1200 and 1100, and they are Bowser x Mario. One guy has a history of not performing worse when he's with a worse player, and better when he's with a better player, so he gets an "a" value of 1.0. The other guy has a history of performing better with people who are both worse and better than him, so he gets an "a" of 0.9. The Mario x Bowser team is known to suck major butt, so it gets a 0.87 value.
Team Elo = (1200 + 1100) * (0.9)(1.0)(0.87)(1 - 0.04) / 2 = 864
So they are a terrible team. But then they turn into Cloud mains:
Team Elo = (1200 + 1100) * (0.9)(1.0)( 1.5)(1 - 0.04) / 2 = 1490
Again, I'm just making this up, so it's not sound. We do have the data, and could come up with an equation that fits well with past results, and then continue to tweak it with future results and its ability to predict the outcome. It's more work than the straightforward "registered team" approach, which would be an easy and quick solution, but it makes for a more malleable option... O_O
tl;dr: With a big enough set of matches, you can introduce 2 new variables (or 3, depending on your starting point) to each *player* that effectively factors in their "synergy" and "skill" in a team setting.
I don't think we should use it, since it's based on a ton of Halo 2 data because 1. we don't have that volume of data, and 2. Smash is a different game. However, it highlights that there may be a more interesting and fun way to come up with a good solution on our own. We could use their suggestion and come up with our own variables to account for variances. One example is character compliment, where two different characters work well together and don't impede each other, which is someone that doesn't exist in Halo 2 at all, and the synergy was all player-based.
There was a CrazyMonkeyKing who said something along the lines of, "All Indians are homophobic assholes" in the twitch chat after mistaking me for an Indian. If this is the same racist person, I'm amused more than anything else about his cry for maturity.
FYI, I will be there and will be taking names down! Doesn't matter if you're Jezmo, Jaice, or a non-QLDer (in which case, you don't matter because QLD is _theBestState). I will win. :-O
Shitashi said: Everything about this event was amazing I want to thank everyone i spoke to individually but I don't have the time and am scared I'll leave people out so for now I'm just going to leave this
That photo creates so many questions...