EPT[Patch 4.20] Blaze It ◯０o。(ー。ー)y~~ - Pag…
| Forum Index > LoL General |
|
chalice
United States1945 Posts
| ||
|
Gahlo
United States35117 Posts
On December 09 2014 00:09 chalice wrote: however, i did hear a rumor that Thorin was studying data science under the tutelage of Stephen A. Smith and Skip Bayless. Thorin might as well be the Skip Bayless of League. | ||
TheYango
United States47024 Posts
On December 08 2014 21:46 Goragoth wrote: I don't get it. You are saying that we already know how to construct great statistics and the people currently trying to come up with statistics are creating bad ones? I thought the whole point was that the field of LoL statistics is still very underdeveloped? In which case, of course any stats we come up with will suck for a long time, but that's not a reason to give up. That's like deciding to become an artist, drawing one picture, saying "well this sucks" and giving up. You have to go through the shitty to get to the good, that's how progress is made. It's an iterative process and we are still at the beginning. Also, while MMR is a great measure for soloqueue performance, certainly something else is needed for analysing players in pro teams. Since there we have many games with the same players winrate doesn't mean a whole lot (we've seen a number of bad players getting carried on good teams and good players getting dragged down by bad teams). When a manager is looking to poach a player for a specific position they will want stats to look at to help them make an objective decision. I'm saying that people are trying to ascribe stats to things before we even have a systematic understanding of how anything works. We don't even have a qualitative model for the game, trying to make a quantitative one is trying to build an automobile before we've even invented the wheel. On December 08 2014 22:09 Sufficiency wrote: The general public does take longer to accept certain scientific theories, often due to their lack of training in scientific method and general unfamiliarity with nee concepts. But if you are at least lightly trained on the subject matter it's not hard to see the merits of someone else's theory by yourself. "Decades" is a gross exaggeration only for the untrained eyes. I should have said "a decade"--I was more pointing to the fact that you have to study the fundamentals of a particular science for like 6-8 years before you even start legitimately doing research science in any real capacity, and even then, how meaningful someone's PhD thesis is a body of research depends on the institution, the science in question, and many other things. On December 08 2014 23:44 cLutZ wrote: Actually, Yango, Bill James (Mr. SABRmetrics in baseball) and Aaron Schatz (Creator of DVOA, and I used a bad term there. I didn't mean sports writers referring to journalists, but more generally "someone who writes about sports"--the implication being that they've studied the game in depth and think about it more or less full time. | ||
|
cLutZ
United States19573 Posts
On December 09 2014 00:20 TheYango wrote: I'm saying that people are trying to ascribe stats to things before we even have a systematic understanding of how anything works. We don't even have a qualitative model for the game, trying to make a quantitative one is trying to build an automobile before we've even invented the wheel. I should have said "a decade"--I was more pointing to the fact that you have to study the fundamentals of a particular science for like 6-8 years before you even start legitimately doing research science in any real capacity, and even then, how meaningful someone's PhD thesis is a body of research depends on the institution, the science in question, and many other things. 1. No one knows what you mean. You probably don't even know what your saying here. 2. The whole point of advanced stats is that they have, on a great number of occasions, bucked the "qualitative" traditional wisdom. Things like "run to pass" became "pass to run"; RBIs are now known to be a useless stat; Fielding Independent pitching is an incredible revolution that turns pitching evaluation on its head; and the 3 point shot was thought to be a silly thing, now its considered one of the most efficient shots in the game. | ||
GrandInquisitor
New York City13113 Posts
The same is not true for League. For starters, there's much more information needed to describe any particular gamestate. A moment in soccer can be accurately and mostly summarized by time remaining + current score. A moment in League would require KDA breakdowns of every champion, towers taken, dragons taken, vision control, and so on. Second, even if we grossly oversimplified by just looking at historical games with the same kill differential / gold lead at X minutes, there still just isn't enough of a sample in the pro scene to conduct meaningful analysis. The reason this is important is because it's one of the only ways to defeat the correlation/causation issue. Guess what's really correlated with winning? Taking dragons. How is that going to help you, though, unless you can say confidently if you're supposed to be taking a dragon or taking a tower? Teams that win are naturally going to take more dragons. It's like NFL analysts that talk about a team's winning record when X running back gets Y carries; he's getting so many carries because they give him the ball so much during winning situations when they are running down the clock. Anyway, the gist of it is, people trying to tell you they're compiling advanced stats for League are either a revolutionary leap ahead of all of the rest of us in terms of League knowledge, or bullshitting because they read the first two chapters of Moneyball and want to look impressive. | ||
|
TheYango
United States47024 Posts
On December 09 2014 00:31 cLutZ wrote: 2. The whole point of advanced stats is that they have, on a great number of occasions, bucked the "qualitative" traditional wisdom. Things like "run to pass" became "pass to run"; RBIs are now known to be a useless stat; Fielding Independent pitching is an incredible revolution that turns pitching evaluation on its head; and the 3 point shot was thought to be a silly thing, now its considered one of the most efficient shots in the game. But that qualitative wisdom had to exist in the first place for you to start applying statistics to things in the first place. You have to have general concepts of how the game "works" (and that's easier in sports where you have more isolated events). Even of some of them are wrong, you correct the model and move forward. You can't blindly apply statistics to things where no model exists because we don't have a basic framework for the game in the first place. | ||
|
cLutZ
United States19573 Posts
On December 09 2014 00:36 TheYango wrote: But that qualitative wisdom had to exist in the first place for you to start applying statistics to things in the first place. You have to have general concepts of how the game "works" (and that's easier in sports where you have more isolated events). Even of some of them are wrong, you correct the model and move forward. You can't blindly apply statistics to things where no model exists because we don't have a basic framework for the game in the first place. You have yet to tell me what we need to "know" in order to make such a model. Which, by the way, I think would be hard to do (and fairly useless) because of how quick patch cycles are. | ||
|
nafta
Bulgaria18893 Posts
| ||
|
Kupon3ss
時の回廊10066 Posts
| ||
Chexx
Korea (South)11232 Posts
On December 09 2014 00:41 Kupon3ss wrote: just use riot-approved fantasy stats imo, the fantasy point is clearly the advanced algorithm that is sanctioned by the creator of the game to gauge the relative worth of players on roles. shouldnt you be busy with the DotA site? | ||
|
chalice
United States1945 Posts
On December 09 2014 00:31 GrandInquisitor wrote:Anyway, the gist of it is, people trying to tell you they're compiling advanced stats for League are either a revolutionary leap ahead of all of the rest of us in terms of League knowledge, or bullshitting because they read the first two chapters of Moneyball and want to look impressive. who are the people making these claims that you and fakeyango are referring to though? i am legit unaware of anyone who has done this kind of stuff and i would be grateful if one of you could direct me to where i could find more information about them. even if they are clueless kids with "shitty models", i think seeing how they are approaching things would still be educational. | ||
|
Purge
Canada372 Posts
Just checking out the TL forums for the first time in a while and this conversation is really intriguing. I am just curious though, Anyone know how the dragons math out compared to relative gold leads right now? I used to just compare the difference between gold earned by teams against the losing teams score as a percentage and that usually gave me a good indicator of relative strength, but now thats not possible because its a bit harder to quantify and is a fair bit abstract. I know the Also so on the topic of this who are the people making these claims that you and fakeyango are referring to though? i am legit unaware of anyone who has done this kind of stuff and i would be grateful if one of you could direct me to where i could find more information about them. even if they are clueless kids with "shitty models", i think seeing how they are approaching things would still be educational. Its sounds so much like a catch 22 that its unbelieveable. Noone is going to tell you their advanced stats because as it stands anyone with the technical knowhow is going to be getting snapped up by a team. By that rationale shouldnt people just start at least giving ideas an ear instead of dismissing them out of hand? And what do you guys think of Viktor on 4.20? I honestly think he is sleeper OP and completely competitively viable just based on his recent rework, but Im not sure in what context. All the arguments against him look the same to the ones given by people who discarded Syndra as a competitive pick last year. | ||
|
Sufficiency
Canada23833 Posts
On December 09 2014 00:50 chalice wrote: who are the people making these claims that you and fakeyango are referring to though? i am legit unaware of anyone who has done this kind of stuff and i would be grateful if one of you could direct me to where i could find more information about them. even if they are clueless kids with "shitty models", i think seeing how they are approaching things would still be educational. Did that model take 10 years to develop? No? It must be garbage then. | ||
|
Slusher
United States19143 Posts
ex. (with 1 death) a solo kill is 1 kda, and a pick with your entire team is .2 kda not 100% indicative of skill, but it told you something about a player, i.e. it was not at all surprising ZionSpartian led the NA lcs in this statistic | ||
|
Slusher
United States19143 Posts
| ||
|
Sufficiency
Canada23833 Posts
On December 09 2014 01:52 Slusher wrote: I kinda liked Jalbert's (Kills or assists/assists on particular kill/assist)/deaths ex. (with 1 death) a solo kill is 1 kda, and a pick with your entire team is .2 kda not 100% indicative of skill, but it told you something about a player, i.e. it was not at all surprising ZionSpartian led the NA lcs in this statistic Kill sharing, right? I think it makes a lot of sense and probably reduce game length dependency like KDA has (see my plot on the previous page). The reason why I am personally not too interested is because the Riot API does not capture this information and it is probably impossible to calculate. So I am stuck with something that is a function of K, D, and A. At the end off the day it is probably not too bad once I make the appropriate adjustments. | ||
|
Slusher
United States19143 Posts
idk my perception is perhaps skewed by an ability to get carried? which a lot of us have surmised in the past is part of the skill set to reach high rank. I wonder what statistics would show if you could find lowest Kda on the winning team per win, plotted against elo. again prob not possible with api sadly | ||
|
chalice
United States1945 Posts
i'm not including any of the full timeline data in the initial parse of the data i collected (although i did download it all), but i'm pretty sure it would be relatively simple to calculate this jalbert kill share stat since they include a list of assisting participants for each event where applicable. i don't have any idea how valuable this KDA variation actually is, but this is definitely along the lines of the suggestions that i was looking for and i appreciate anything that might give me an excuse to start working with the timeline data. | ||
|
Sufficiency
Canada23833 Posts
On December 09 2014 02:44 chalice wrote: i think part of the off-hand dismissal of the analytic possibilities stems from a lack of understanding of just how much data is actually available from the api. although there are certainly a lot of complicated variables that you have to navigate through, because of the electronic nature of the sport riot is able to provide you with pretty much the equivalent of the advanced SportVU player tracking cameras that NBA teams are just now beginning to use and cost millions, for free with the timeline data you can get from the api. i'm not including any of the full timeline data in the initial parse of the data i collected (although i did download it all), but i'm pretty sure it would be relatively simple to calculate this jalbert kill share stat since they include a list of assisting participants for each event where applicable. i don't have any idea how valuable this KDA variation actually is, but this is definitely along the lines of the suggestions that i was looking for and i appreciate anything that might give me an excuse to start working with the timeline data. You have a good point. But honestly I haven't even tried the new match history yet because there is already too muvh data for me to analyse from the vanilla history. | ||
|
JimmiC
Canada22817 Posts
| ||
| ||
