|
On March 28 2014 11:11 TheYango wrote:Show nested quote +On March 28 2014 11:06 Ketara wrote:
But what it does do is add a little bit of reliability to what we've said. I didn't put it in the guide because I thought it was a definitive set of statistics on Lux matchups. If it came up with a matchup that looked wildly different to my own experience, I'd be pretty confident in calling the statistic model wrong. But since said statistics and our game experience more or less agree, I think it's good to include because they add weight to each other. So then where does the value come from If it holds zero weight when it differs from actual experience?
The value comes from them agreeing. It adds credibility to my experience by showing that this experience is not wildly different from everybody elses.
If it were different, I'd be confident in calling it wrong, but I'd try to look for some reason why the statistics showed that. For example, Suffs first set of data for Lux (before he realized there was a problem with it) showed Lux absolutely beating the shit out of Kassadin.
This is a particular matchup that Dai and I have discussed a lot, because when I was spamming Lux games in season 3 and grinding through Bronze, Silver and Gold, I played against Kassadin a lot, and I did beat the shit out of him. Dai however always told me that Kassadin was more or less a hard counter to Lux. I was skeptical about this, but as I got higher up on the ladder, Kassadin players started, frankly, not playing like idiots.
So I can say with some confidence now that the Lux vs. pre rework Kassadin matchup was a very difficult one. However, I can see why the statistics data would say otherwise, because it's a volatile matchup and most of the games used to model said matchup are going to be low level games. I have enough knowledge of the matchup to know that if a Lux who doesn't know what she's doing goes up against a Kassadin who doesn't know what he's doing, Lux will have a much better time than if both players are playing it correctly.
The point is that it's not useless data. It's not as good as game experience and understanding, but it is not useless. I don't think I've ever seen Suff try to claim that it's more credible data than what it is, a simple curiosity or possible reference material. And this is why I don't understand the animosity.
|
That's not data agreeing with you, that's the data not directly contradicting your experience, that is to say under sufficiency's example
I just solved this math problem and the answer I got was 42. I asked my friend, he told me he also got a postive number using a different method. Which proves that my method and answer are correct and relevant
|
United States47024 Posts
On March 28 2014 11:21 Ketara wrote:
The point is that it's not useless data. It's not as good as game experience and understanding, but it is not useless. I don't think I've ever seen Suff try to claim that it's more credible data than what it is, a simple curiosity or possible reference material. And this is why I don't understand the animosity.
Other than the title being "champion counters".
I'll admit that part of this is me taking this out on Sufficiency for more broad issue I take with the idea of analyzing drafting in this manner.
|
As long as you come at it with the understanding that it's just blanket numbers for 1v1 comparisons and skewed based on
A - Player league
B - The other 8 champions on the team
C - Individual game bullshittery
I think it's fine. In fact, I think it's great. I wish we did shit like this more often.
|
|
|
It took me a moment to realize that was a 45-degree line and not a trendline.
As for the broader discussion, I'd like to thank Sufficiency for the clearly significant amount of time that he's put into providing this data for the community. While I'm sure that there are methodological points to be discussed at length, I think that we're being a bit too confrontational about it. It looks like neither side is conclusively persuading the other, and it poisons the well for the rest of us.
On an unrelated note, what do you do as a jungler when you find your 2nd buff stolen? With the trinket changes, it's been happening more frequently, and I'm not sure how to proceed, beyond pinging danger and saying in chat that it's gone, be careful. Having to just farm small camps while watching the enemy jungler snowball lanes is really frustrating.
|
Baa?21248 Posts
On March 28 2014 11:21 Ketara wrote:Show nested quote +On March 28 2014 11:11 TheYango wrote:On March 28 2014 11:06 Ketara wrote:
But what it does do is add a little bit of reliability to what we've said. I didn't put it in the guide because I thought it was a definitive set of statistics on Lux matchups. If it came up with a matchup that looked wildly different to my own experience, I'd be pretty confident in calling the statistic model wrong. But since said statistics and our game experience more or less agree, I think it's good to include because they add weight to each other. So then where does the value come from If it holds zero weight when it differs from actual experience? The value comes from them agreeing. It adds credibility to my experience by showing that this experience is not wildly different from everybody elses. If it were different, I'd be confident in calling it wrong, but I'd try to look for some reason why the statistics showed that. For example, Suffs first set of data for Lux (before he realized there was a problem with it) showed Lux absolutely beating the shit out of Kassadin. This is a particular matchup that Dai and I have discussed a lot, because when I was spamming Lux games in season 3 and grinding through Bronze, Silver and Gold, I played against Kassadin a lot, and I did beat the shit out of him. Dai however always told me that Kassadin was more or less a hard counter to Lux. I was skeptical about this, but as I got higher up on the ladder, Kassadin players started, frankly, not playing like idiots. So I can say with some confidence now that the Lux vs. pre rework Kassadin matchup was a very difficult one. However, I can see why the statistics data would say otherwise, because it's a volatile matchup and most of the games used to model said matchup are going to be low level games. I have enough knowledge of the matchup to know that if a Lux who doesn't know what she's doing goes up against a Kassadin who doesn't know what he's doing, Lux will have a much better time than if both players are playing it correctly. The point is that it's not useless data. It's not as good as game experience and understanding, but it is not useless. I don't think I've ever seen Suff try to claim that it's more credible data than what it is, a simple curiosity or possible reference material. And this is why I don't understand the animosity.
How did "stats" contribute any value at all to your scenario? It was literally your own (worthless) stats being discarded when you realized that you're bad. Statistics added no positive value at all towards your understanding of the game.
Bad stats/math/theorycraft (aka almost all stats/math/theorycraft that get trotted out on this subforum) is worse than just keeping your mouth shut and not bringing out stupid meaningless data.
|
Thanks for the insults Cheep. I'm sure we can all add it to the list of posts you shouldn't be allowed to get away with.
|
On March 28 2014 10:02 Sufficiency wrote:
EDIT: let me give you a more serious response. Words are that you are an economist. What you just said to me is like I just ask you "WTF WHY DO PEOPLE RESPOND TO INCENTIVES?!?!?!".
I am serious. The matter of taking the log is really fundamental. It's like the economic principle that people respond to incentives and everything has a cost.
No. You take the log when it makes sense to do so based on your model. The reason you use the logistic distribution is because you believe that the data fits it. Because you have a binary output and non-binary inputs. There is no more or less of it than that. You do not take the log because it is axiomatic.
A linear regression takes the form y=xB+e where y is a vector of your dependent variables, x is your matrix of dependent variables, and B is the vector of coefficients.
We can note simply that our estimate of B is equal to (x'x)^-1x'y by simple linear algebra. We note similarly by linear algebra that if we take a linear transformation of x we have predictable effects on B. If you double x then you halve B. But the interpretation of the effect is still the same.
Almost any transformation of a binary variable will have similar results. Because even if you square or log* or whatever, the same transformation on the same values produces the same effect each time. With only two values all you're doing is changing the value of B and so its interpretation. But you're not actually changing the underlying statistics. You've written a model which is is a simple transformation that doesn't effect the underlying statistics away from linear then claimed its logistic.
Literally all you have to do is filter games with ashe on one side then run a regression y=xB+e. Where y is the outcome of the game, and x is the vector of other champions on the other side plus a constant. Its not perfect because there are no interaction terms, but as stated earlier we ignore interaction terms because you don't have enough data or enough computer.
*though you shouldn't do this with binary because log(0) is undefined
On March 28 2014 10:47 TheYango wrote:
There would be a lot more meaningful discussion if you'd actually tried to model something to measure relative champion advantage. For example developing an aggregate statistic for laning advantage that is a combination of gold lead at 10 minutes, direct kills against your opponent, overall kill contribution, creep score, etc. A quantitative model for something that you have a qualitative understanding of. At that point we might discuss the validity of the aggregate statistic in conjunction with games played.
This kind of analysis would actually have some more real applicability. Once you use the data to establish the validity of your statistics based on known outcomes (i.e. these are games where we can definitively say player A won the lane, and the statistic matches that), we can use this to address the gray area cases where qualitative analysis fails (e.g. champion A gets behind champion B on CS but typically gank assists better and can gank other lanes more easily--who "wins"?).
A winrate table doesn't do anything.
You have it the other way around. A winrate table is best. This is why MMR is used for matchmaking and values such as "gold advantage" or "CS" aren't. The only reliable information is game outcome.
On March 28 2014 11:05 Sufficiency wrote:Show nested quote +On March 28 2014 10:50 Kupon3ss wrote:
But its a winrate table that combines the aggregate of everything from the equivalent of little league to semi-pro games with 0 rigor or segregation of results operating under a model in which each champion's winrate is extrapolate through all levels of play and possible matchups. How could it possibly be wrong, its got millions of datapoints and a minuscule P-value.
PS: Did you know that the P-value of a test that evaluates whether or not Blue Side counters Purple Side in ARAM is 0? Yes. If you do it properly, you can clearly show there blue side has an advantage. This is in ARAM as well as on Summoners' Rift.
But you need theory to suggest why blue side would have an advantage in ARAM. Theory exists for summoners rift (game is asymmetric), theory exists for champions (counters exist and should win more). Why for ARAM?
I mean lets not quibble, the ability to find "significant" effects in things that we have no reason to believe are significant is a problem.
|
On March 28 2014 11:19 TheYango wrote:Show nested quote +On March 28 2014 11:12 cLutZ wrote:
I dont really want to get into the Math Shitstorm, but I do know something about baseball stats. And what we know about baseball stats is that before people looked at the new, so called "advanced", stats they were wrong about like 1/2+ of the stats they thought were important for winning. So, just FYI, maybe avoid that analogy in the future. That's exactly the point I was trying to get at, though and precisely WHY I used the baseball analogy.
Well it doesnt really work. Because people used (and some old people still do) shitty stats like RBI and Pitcher Wins because they ignored math, at least Suff is trying math and acknowledging the possible flaws.
Baseball people are like the worst.
|
+ Show Spoiler + Heimerdinger,Gangplank, and Shaco coming back after All Stars won vs Masters
|
On March 28 2014 12:49 LightningStrike wrote:+ Show Spoiler + Heimerdinger,Gangplank, and Shaco coming back after All Stars won vs Masters
The first one is OP right now, no big surprise there.
|
On March 28 2014 12:49 LightningStrike wrote:+ Show Spoiler + Heimerdinger,Gangplank, and Shaco coming back after All Stars won vs Masters
"It's legit, I swear. I saw it on OGN!"
|
Baa?21248 Posts
On March 28 2014 12:04 Ketara wrote:
Thanks for the insults Cheep. I'm sure we can all add it to the list of posts you shouldn't be allowed to get away with.
No one is insulting anyone (well maybe Sufficiency is insulting himself), but you guys really need to learn not to get so butthurt everytime someone disagrees with your (incorrect) opinions.
|
problem i have with people posting stats blindly is they don't put it into context. it's easy enough to post numbers and do number crunching, but blindly doing so without any context can lead to this kind of bullshit "x champ counters y champ, the end".
|
Bearded Elder29905 Posts
Does anyone noticed that his match history on lolking.net is not uploading? I mean, points, league, ranked stats works fine but when I do click match history it doesn't show my 3 recent games that were played yesterday and 2 days ago.
|
it should be obvious that if you're looking to gain any kind of deeper understanding of the game that champion winrate statistics are meaningless. even if the model was perfect all you'd end up with is a glorified balance whine, no reason to waste a bunch of time arguing about the statistical rigors of a math based championselect.net.
like i don't see sufficiency really making any grand claims about the importance and meaning of his numbers to warrant quite this amount of scrutiny.
|
On March 28 2014 11:16 Sufficiency wrote:Show nested quote +On March 28 2014 11:11 TheYango wrote:On March 28 2014 11:06 Ketara wrote:
But what it does do is add a little bit of reliability to what we've said. I didn't put it in the guide because I thought it was a definitive set of statistics on Lux matchups. If it came up with a matchup that looked wildly different to my own experience, I'd be pretty confident in calling the statistic model wrong. But since said statistics and our game experience more or less agree, I think it's good to include because they add weight to each other. So then where does the value come from If it holds zero weight when it differs from actual experience? I just solved this math problem and the answer I got was 42. I asked my friend, he told me he got 42 as well using a different method. Unfortunately his answer held zero weight.
You must be aware that you can get correct answers through irrelevant methodology all the time in math right? So it in fact doesn't add any weight at all.
On March 28 2014 14:21 chalice wrote:
like i don't see sufficiency really making any grand claims about the importance and meaning of his numbers to warrant quite this amount of scrutiny.
It was originally "this stuff is cool but you shouldn't read too much into it because on a fundamental level you can't read too much into it." This seems to have struck a nerve with our flagbearers of generation wuss.
|
On March 28 2014 14:21 chalice wrote:
it should be obvious that if you're looking to gain any kind of deeper understanding of the game that champion winrate statistics are meaningless. even if the model was perfect all you'd end up with is a glorified balance whine, no reason to waste a bunch of time arguing about the statistical rigors of a math based championselect.net.
like i don't see sufficiency really making any grand claims about the importance and meaning of his numbers to warrant quite this amount of scrutiny.
And this is the point where the warwick support analogy reaches the "She's not saying it's op, just relax" stage.
|
Czech Republic11293 Posts
I find sufficiency's statistic stuff fun and also somewhat useful. I often don't know how favourable/unfavourable lanes are and when the difference in winrate is pretty big it can indicate that.
|
|
|
|
|
|
EPT![[image loading]](http://www.team-dignitas.net/uploads/tinymce/images/rr-lcs_vs_soloq-winrate3w.png)
