|
I don't think a projectile you can't flash over exists.
|
I'm not really a fan of the approach Riot takes with Maokai's ult. A flat max duration for it would be incredibly meta dependant, as against assassins something like 5-6s would probably be more than enough if you want to peel for a squishy. On the other hand, with the current fights it'd be much more limited, and even moreso in lane when gank attempts can last awhile. Nerfing W at later levels is not surprising, and trying to reduce his tankiness a bit too, but I find that approach awkward and too limiting.
Morgana's values on Black Shield have always been pretty high, especially for a shield. Probably because it only blocks magic damage but, if you wanted to make it stronger to compensate for it being situational, it also means that there are less means of breaking it if what's important is the cc negation part, and it's also stronger than most shields against magic damage.
Fwiw it absorbs around 115 magic EHP at level 1 (assuming 25 MR on a squishy with some MR runes against MPen marks and 0 AP on Morgana), which is more than most spells, meaning until level 4 (or possibly 3) it's probably going to require 2 spells to destroy it before you can start applying cc. Reducing its value at early levels should be fine.
|
On August 06 2014 09:01 Lmui wrote:Show nested quote +On August 06 2014 04:35 swim224 wrote:On August 06 2014 03:21 739 wrote:
Short question, because I was always wondering :
Are you possible to flash over projectiles? I mean are you able to flash on top of jinx's ult when it's going straight forward into you? Not sure about jinx ult because it's pretty long, but you can flash over the vast majority of projectiles. You can flash over any projectile which is shorter than flash range. Some examples that I've personally flashed:: Elise Cocoon Ezreal Q Syndra stun
Flashing Ali W is funny!
On August 06 2014 08:47 IMoperator wrote:
More nerfs on the PBE, for mao/elise/morg.
So does riot's whole balance philosophy revolve around nerfing whatever is used in LCS?
They have an interesting way of nerfing Morg E. There's not much counterplay if you force the other support to break it, because the ad won't have enough magic damage. Why is it so hard to make the duration scale with level? Or make it so that it lasts for shorter duration on teammates.
|
i wish thgey would just start setting the most used LCS champions as a baseline and then design around that.
Right now they just nerf the most played by altering champion defining abilities, which leads to champions being too similar, which leads to the ones with simply better numbers will be played.
Not a fan of this whole thing.
|
On August 06 2014 08:53 Goumindong wrote:Show nested quote +On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it
You're being sufficiently vague throwing around the mathematical terms in that first paragraph that I (and I graduated with a minor in math) have no idea what you mean. (I don't really think that discussion should occur, being that it's essentially an argument of whether ELO hell exists or not. I don't care one way or another, but I would like it if you clarified what you meant mathematically.)
But you're right about the fact that it has no impact what value p has or how you derive it.
|
On August 06 2014 08:53 Goumindong wrote:Show nested quote +On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it
I'm not having an issue with people in ranked games, to which they said they kept track of how many it was before it got too frustrating to do so. Then again, instead of dealing with actual data you'd probably just dismiss it out of hand. For the record, I was pointing out your misunderstanding for clarity's sake, not that I thought it makes a significant difference.
|
On August 06 2014 09:31 sylverfyre wrote:Show nested quote +On August 06 2014 08:53 Goumindong wrote:On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it You're being sufficiently vague throwing around the mathematical terms in that first paragraph that I (and I graduated with a minor in math) have no idea what you mean. (I don't really think that discussion should occur, being that it's essentially an argument of whether ELO hell exists or not. I don't care one way or another, but I would like it if you clarified what you meant mathematically.) But you're right about the fact that it has no impact what value p has or how you derive it.
It was an aside about discussing ELO hell because Gahlo is essentially saying that ELO hell can exist because you can totally just get unlucky. Specifically if there exists a population of players which are better than their ELO would suggest (say by constantly dodging dropping MMR when that was a thing) then games can get easier as you win and harder as you lose(rather than the other way around) if the probability that you will run into a smurf is decreasing as you increase in MMR. This probability is the ratio of the distributions of "properly placed players" and "smurfs", both of which are normal distributions. In order to actually make an ELO Hell there almost must also be a sufficient density of smurfs because the probability decrease in smurfs has to counteract the difficulty in playing better opponents normally.
The probability has to be decrease so the derivative of the ratio of the weighted distributions has to be negative. Going backwards from there shouldn't be that difficult if you knew what was being discussed. This is one of the reasons that Riot stopped having MMR penalties for dodging, because if enough people dodge games consistently it can create that effect.
But this doesn't exist as we know it in league. ELO hell aint real.
On August 06 2014 09:34 Gahlo wrote:
I'm not having an issue with people in ranked games, to which they said they kept track of how many it was before it got too frustrating to do so. Then again, instead of dealing with actual data you'd probably just dismiss it out of hand. For the record, I was pointing out your misunderstanding for clarity's sake, not that I thought it makes a significant difference.
On this case i will absolutely dismiss it out of hand because biased observers cannot generate reliable data. When they play a game they have no way to assess the raginess of the other team except ask the other team. Which A) they're unlikely do do and B) Even if they do they won't get a response that they can compare to their own data and C) even if they do and they get a response that they can compare to their own data [which they can't] they still don't know if they're the problem.
The only way to get reliable data is to have an unbiased observer watch the replays from both sides including chat and judge the number of ragers on each side. But nobody has time for that shit.
I mean look i've probably watched games from like 10-12 different people who have said "i am totally in ELO hell and am playing just fine" and every time i say "No, its definitely you. You're making lots of mistakes and probably raging" and i go and watch their games and yup, they're making mistakes and throwing and raging and contributing to losses. Sure 40-50 games for 10 people isn't much of a sample size but frankly i don't think anyone has done this type of thing and been like "yea you just totally get screwed in all your games"
|
On August 06 2014 08:53 Goumindong wrote:Show nested quote +On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it
Sounds pretty shitty to me.
|
On August 06 2014 10:15 Goumindong wrote:Show nested quote +On August 06 2014 09:31 sylverfyre wrote:On August 06 2014 08:53 Goumindong wrote:On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it You're being sufficiently vague throwing around the mathematical terms in that first paragraph that I (and I graduated with a minor in math) have no idea what you mean. (I don't really think that discussion should occur, being that it's essentially an argument of whether ELO hell exists or not. I don't care one way or another, but I would like it if you clarified what you meant mathematically.) But you're right about the fact that it has no impact what value p has or how you derive it. It was an aside about discussing ELO hell because Gahlo is essentially saying that ELO hell can exist because you can totally just get unlucky. Specifically if there exists a population of players which are better than their ELO would suggest (say by constantly dodging dropping MMR when that was a thing) then games can get easier as you win and harder as you lose(rather than the other way around) if the probability that you will run into a smurf is decreasing as you increase in MMR. This probability is the ratio of the distributions of "properly placed players" and "smurfs", both of which are normal distributions. In order to actually make an ELO Hell there almost must also be a sufficient density of smurfs because the probability decrease in smurfs has to counteract the difficulty in playing better opponents normally. The probability has to be decrease so the derivative of the ratio of the weighted distributions has to be negative. Going backwards from there shouldn't be that difficult if you knew what was being discussed. This is one of the reasons that Riot stopped having MMR penalties for dodging, because if enough people dodge games consistently it can create that effect. But this doesn't exist as we know it in league. ELO hell aint real. Show nested quote +On August 06 2014 09:34 Gahlo wrote:
I'm not having an issue with people in ranked games, to which they said they kept track of how many it was before it got too frustrating to do so. Then again, instead of dealing with actual data you'd probably just dismiss it out of hand. For the record, I was pointing out your misunderstanding for clarity's sake, not that I thought it makes a significant difference. On this case i will absolutely dismiss it out of hand because biased observers cannot generate reliable data. When they play a game they have no way to assess the raginess of the other team except ask the other team. Which A) they're unlikely do do and B) Even if they do they won't get a response that they can compare to their own data and C) even if they do and they get a response that they can compare to their own data [which they can't] they still don't know if they're the problem. The only way to get reliable data is to have an unbiased observer watch the replays from both sides including chat and judge the number of ragers on each side. But nobody has time for that shit. I mean look i've probably watched games from like 10-12 different people who have said "i am totally in ELO hell and am playing just fine" and every time i say "No, its definitely you. You're making lots of mistakes and probably raging" and i go and watch their games and yup, they're making mistakes and throwing and raging and contributing to losses. Sure 40-50 games for 10 people isn't much of a sample size but frankly i don't think anyone has done this type of thing and been like "yea you just totally get screwed in all your games"
Good job putting words into my mouth. That's definitely not my definition of ELO hell. Then again, it's probably too plebian for you to understand.
|
On August 06 2014 09:31 sylverfyre wrote:Show nested quote +On August 06 2014 08:53 Goumindong wrote:On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it You're being sufficiently vague throwing around the mathematical terms in that first paragraph that I (and I graduated with a minor in math) have no idea what you mean. (I don't really think that discussion should occur, being that it's essentially an argument of whether ELO hell exists or not. I don't care one way or another, but I would like it if you clarified what you meant mathematically.) But you're right about the fact that it has no impact what value p has or how you derive it.
For the record I don't really quite understand most of his stuff either.
Goumindong is really good at BSing buzz words but I am fairly sure he doesn't actually know as much as he thinks he does. I am certain that he is a good economist; he is not a statistician.
|
Look, this is a dumb conversation. especially when one side refuses to acknowledge that there can even be one person stuck at a lower rank than he/she belongs. if its truly random whether or not a player is a rager/afker there are some people who have that rager on the other team disproportionately more than on their team, and some that have it disproportionately on their team compared to the other team. there are always going to be outliers, an argument against outliers is not "well on average it shouldn't matter".
|
Can someone explain to me the Orianna vs. Zed matchup as Ori in mid lane? Just got done playing a game and I thought Ori was a good person to put against Zed, as you just ball yourself when he ults in and just farm with the ball. He was ganked twice mid (died once, blew flash to escape the second gank) and still came out ahead by 20-30 cs and just snowballed that advantage into the other lanes. We traded kills (I died early, I got him back when he tower dove too deep and ignite killed him). Was my mistake in that I should have taken Exhaust mid to counter the dive along w/ the death mark?
|
Certain topics are blacklisted from LoL General Discussion and they include:
"Elo hell"
Also, this argument is stupid. Yes, luck is a factor. Maybe there's one guy out there who happens to have the worst luck ever and consistently gets bad team mates and is stuck at a ranking lower than he deserves. Chances are, that guy isn't you or anyone you know. Who cares?
|
On August 06 2014 10:49 Gahlo wrote:Show nested quote +On August 06 2014 10:15 Goumindong wrote:On August 06 2014 09:31 sylverfyre wrote:On August 06 2014 08:53 Goumindong wrote:On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:On August 05 2014 14:47 Osmoses wrote:
That doesn't actually apply in this instance. How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers. In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it You're being sufficiently vague throwing around the mathematical terms in that first paragraph that I (and I graduated with a minor in math) have no idea what you mean. (I don't really think that discussion should occur, being that it's essentially an argument of whether ELO hell exists or not. I don't care one way or another, but I would like it if you clarified what you meant mathematically.) But you're right about the fact that it has no impact what value p has or how you derive it. It was an aside about discussing ELO hell because Gahlo is essentially saying that ELO hell can exist because you can totally just get unlucky. Specifically if there exists a population of players which are better than their ELO would suggest (say by constantly dodging dropping MMR when that was a thing) then games can get easier as you win and harder as you lose(rather than the other way around) if the probability that you will run into a smurf is decreasing as you increase in MMR. This probability is the ratio of the distributions of "properly placed players" and "smurfs", both of which are normal distributions. In order to actually make an ELO Hell there almost must also be a sufficient density of smurfs because the probability decrease in smurfs has to counteract the difficulty in playing better opponents normally. The probability has to be decrease so the derivative of the ratio of the weighted distributions has to be negative. Going backwards from there shouldn't be that difficult if you knew what was being discussed. This is one of the reasons that Riot stopped having MMR penalties for dodging, because if enough people dodge games consistently it can create that effect. But this doesn't exist as we know it in league. ELO hell aint real. On August 06 2014 09:34 Gahlo wrote:
I'm not having an issue with people in ranked games, to which they said they kept track of how many it was before it got too frustrating to do so. Then again, instead of dealing with actual data you'd probably just dismiss it out of hand. For the record, I was pointing out your misunderstanding for clarity's sake, not that I thought it makes a significant difference. On this case i will absolutely dismiss it out of hand because biased observers cannot generate reliable data. When they play a game they have no way to assess the raginess of the other team except ask the other team. Which A) they're unlikely do do and B) Even if they do they won't get a response that they can compare to their own data and C) even if they do and they get a response that they can compare to their own data [which they can't] they still don't know if they're the problem. The only way to get reliable data is to have an unbiased observer watch the replays from both sides including chat and judge the number of ragers on each side. But nobody has time for that shit. I mean look i've probably watched games from like 10-12 different people who have said "i am totally in ELO hell and am playing just fine" and every time i say "No, its definitely you. You're making lots of mistakes and probably raging" and i go and watch their games and yup, they're making mistakes and throwing and raging and contributing to losses. Sure 40-50 games for 10 people isn't much of a sample size but frankly i don't think anyone has done this type of thing and been like "yea you just totally get screwed in all your games" Good job putting words into my mouth. That's definitely not my definition of ELO hell. Then again, it's probably too plebian for you to understand.
When you say that people saying that "one person won't stop your rise" is the gambler's fallacy what do you expect people to think? And that "there won't be an even distribution of trolls/ragers because no one plays that many games"?
Note that the gambler's fallacy doesn't actually exist in League because winning more legitimately makes the game harder as does losing make the game easier (so long as your performance also doesn't correlate with wins/losses)
|
On August 06 2014 10:51 PrinceXizor wrote:
Look, this is a dumb conversation. especially when one side refuses to acknowledge that there can even be one person stuck at a lower rank than he/she belongs. if its truly random whether or not a player is a rager/afker there are some people who have that rager on the other team disproportionately more than on their team, and some that have it disproportionately on their team compared to the other team. there are always going to be outliers, an argument against outliers is not "well on average it shouldn't matter".
Sure there can be outliers but for any "outlier" it significantly more likely they're the problem than they're unlucky. That is to say that outliers may be theoretically possible but good luck finding them
|
On August 06 2014 10:51 PrinceXizor wrote:
Look, this is a dumb conversation. especially when one side refuses to acknowledge that there can even be one person stuck at a lower rank than he/she belongs. if its truly random whether or not a player is a rager/afker there are some people who have that rager on the other team disproportionately more than on their team, and some that have it disproportionately on their team compared to the other team. there are always going to be outliers, an argument against outliers is not "well on average it shouldn't matter".
It should be clear from the get-go that when you have 4 teammates on your side and 5 on the other, you will be less likely to get trolls/nubs/etc on your side provided you are not a troll/nubs/etc yourself. This is common sense and can be explained without all the LLN crap.
The real question is how much difference there is between the two teams just because you constitute one player for one of the two teams. That can be answered using some form of statistical inference, not LLN.
|
On August 06 2014 10:57 Goumindong wrote:Show nested quote +On August 06 2014 10:51 PrinceXizor wrote:
Look, this is a dumb conversation. especially when one side refuses to acknowledge that there can even be one person stuck at a lower rank than he/she belongs. if its truly random whether or not a player is a rager/afker there are some people who have that rager on the other team disproportionately more than on their team, and some that have it disproportionately on their team compared to the other team. there are always going to be outliers, an argument against outliers is not "well on average it shouldn't matter". Sure there can be outliers but for any "outlier" it significantly more likely they're the problem than they're unlucky. That is to say that outliers may be theoretically possible but good luck finding them
thats not how it works mate.
|
On August 06 2014 10:55 Goumindong wrote:Show nested quote +On August 06 2014 10:49 Gahlo wrote:On August 06 2014 10:15 Goumindong wrote:On August 06 2014 09:31 sylverfyre wrote:On August 06 2014 08:53 Goumindong wrote:On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:
[quote]
How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers.
In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:On August 05 2014 14:49 Gahlo wrote:
[quote]
How doesn't it? I doubt anybody plays enough soloque games to qualify for law of large numbers.
In addition, Law of Large #s doesn't apply anyway. Things like coin flips and dice rolls have set outcomes. Who matchmaking pulls out of a hat of thousands from differing size and valued pools has way too many shifting parts. LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it You're being sufficiently vague throwing around the mathematical terms in that first paragraph that I (and I graduated with a minor in math) have no idea what you mean. (I don't really think that discussion should occur, being that it's essentially an argument of whether ELO hell exists or not. I don't care one way or another, but I would like it if you clarified what you meant mathematically.) But you're right about the fact that it has no impact what value p has or how you derive it. It was an aside about discussing ELO hell because Gahlo is essentially saying that ELO hell can exist because you can totally just get unlucky. Specifically if there exists a population of players which are better than their ELO would suggest (say by constantly dodging dropping MMR when that was a thing) then games can get easier as you win and harder as you lose(rather than the other way around) if the probability that you will run into a smurf is decreasing as you increase in MMR. This probability is the ratio of the distributions of "properly placed players" and "smurfs", both of which are normal distributions. In order to actually make an ELO Hell there almost must also be a sufficient density of smurfs because the probability decrease in smurfs has to counteract the difficulty in playing better opponents normally. The probability has to be decrease so the derivative of the ratio of the weighted distributions has to be negative. Going backwards from there shouldn't be that difficult if you knew what was being discussed. This is one of the reasons that Riot stopped having MMR penalties for dodging, because if enough people dodge games consistently it can create that effect. But this doesn't exist as we know it in league. ELO hell aint real. On August 06 2014 09:34 Gahlo wrote:
I'm not having an issue with people in ranked games, to which they said they kept track of how many it was before it got too frustrating to do so. Then again, instead of dealing with actual data you'd probably just dismiss it out of hand. For the record, I was pointing out your misunderstanding for clarity's sake, not that I thought it makes a significant difference. On this case i will absolutely dismiss it out of hand because biased observers cannot generate reliable data. When they play a game they have no way to assess the raginess of the other team except ask the other team. Which A) they're unlikely do do and B) Even if they do they won't get a response that they can compare to their own data and C) even if they do and they get a response that they can compare to their own data [which they can't] they still don't know if they're the problem. The only way to get reliable data is to have an unbiased observer watch the replays from both sides including chat and judge the number of ragers on each side. But nobody has time for that shit. I mean look i've probably watched games from like 10-12 different people who have said "i am totally in ELO hell and am playing just fine" and every time i say "No, its definitely you. You're making lots of mistakes and probably raging" and i go and watch their games and yup, they're making mistakes and throwing and raging and contributing to losses. Sure 40-50 games for 10 people isn't much of a sample size but frankly i don't think anyone has done this type of thing and been like "yea you just totally get screwed in all your games" Good job putting words into my mouth. That's definitely not my definition of ELO hell. Then again, it's probably too plebian for you to understand. When you say that people saying that "one person won't stop your rise" is the gambler's fallacy what do you expect people to think? And that "there won't be an even distribution of trolls/ragers because no one plays that many games"? Note that the gambler's fallacy doesn't actually exist in League because winning more legitimately makes the game harder as does losing make the game easier (so long as your performance also doesn't correlate with wins/losses)
The more this line of discussion goes, it becomes more apparent to me that you aren't even reading my posts or just scimming them and assuming you know what I'm saying. Whether you like it or not, it is possible for a player to get a bunch of shit teammates in a decent sized cluster of game and regardless of that, there is never a guarantee in any form or fashion that the tide would reverse.
|
On August 06 2014 11:08 Gahlo wrote:Show nested quote +On August 06 2014 10:55 Goumindong wrote:On August 06 2014 10:49 Gahlo wrote:On August 06 2014 10:15 Goumindong wrote:On August 06 2014 09:31 sylverfyre wrote:On August 06 2014 08:53 Goumindong wrote:On August 06 2014 08:31 Gahlo wrote:On August 06 2014 05:50 JimmiC wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:
[quote]
LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. The Gahlo doth protest to much, me thinks. http://en.wikipedia.org/wiki/The_lady_doth_protest_too_much,_methinks Why? I'm doing well in soloq lately and my LP splits tell me I should expect to do so for a while. I'm just saying that people can "be unlucky" with ragers and whatnot in soloq. It happens, trying to say it doesn't is ludicrous. On August 06 2014 06:04 Goumindong wrote:On August 06 2014 05:27 Gahlo wrote:On August 06 2014 03:26 Goumindong wrote:
[quote]
LLN applies. The average of the sum of different distributions is the average of the distributions. You will not, over a sufficiently large number of games played, get more or less ragers than anyone else in your same bracket or who is raising/falling. There will be variance but unless you're the problem you will rise because of the LLN. The more variables involved in the test, the higher the sample size required. The most common example of tLoL#s is flipping a coin 100 times and getting around a 50:50 ratio. But that comes with the standards of it being a two sided coin with one heads, the other tails and it's always the same coin or coins exactly like it. Soloque is like rolling 9 multimillion sided dice, where each have different distributions of heads and tails markings on them that can change their sides after rolling while still being determined and constantly swapping in and out hundreds of dice. No. Its true that the less variance between the polled distributions (not variance of the polled distributions) the faster things converge, but soloqueue is not like rolling a 9 multimillion sided dice (and even if it was that wouldn't actually matter) In soloqueue the probability distribution that you get ragers/leavers on your team is Binomial n=4, p = ? the probability that you get ragers/leavers on the other team is Binomial n=5 p = ?. For any p, Binomial n=5 dominates n=4 (in that the probability of x or more success is necessarily higher for all x besides zero). The p has to be the same because when you queue for soloqueue you're pulling from the same distribution of people and while you're doing so without replacement the population is large enough that we can look at it like its a with replacement problem without really any loss of accuracy (and note that the with replacement problem still has Binomial n=5 dominating Binomial n=4 for any population) This isn't a 9 million sided die, its like rolling a d100 9 times for each instance and sometimes we record a success on a 9 or lower and sometimes we record a success on a 10 or lower (depending on the ratio of ragers to non ragers who ar online in your bracket at this moment. This makes the variance of the difference of ragers that are on the enemy team to ragers on your team pretty lowBy the law of large numbers we can know that the sum of your random rage difference will asymptotically be the weighted average of the rage difference distributions. Because Binomial 5 dominates Binomial 4 we know that this number will be negative in all cases (that is we expect in more ragers on the enemy team always) then for every person it is the case that as they play more games they will always have more ragers on the enemy team. Asymptotically we get there pretty fast, a hundred or so games will make the likelihood that you have more ragers on the enemy team only a few %. Two hundred games and its basically zero. The lower variance between the p's the faster we converge to the proper % but this doesn't have much of an effect on how fast we converge to "below zero with high certainty" The only time this doesn't hold true is if something that you're doing is causing people to rage and this something is consistent across your games such that in it actually increases the probability that people rage/afk in your games but not on the enemy team. So we have to look at what is more likely when someone legitimately gets more ragers on his team. Is it more likely that he is supremely unlucky, or is it more likely that he is an asshole that makes other people rage? Answer: its far more likely he is an asshole who makes other people rage. Note that there have been situations in the past which could create ELO hell, but describing that is another long post that I don't want to do and you probably wouldn't understand anyway. But as far as I can tell, Riot fixed that situation, and the other instance in which it can occur doesn't seem like enough of a problem to worry about. edit: The other thing that can be happening is observation bias. When people on their team rage and AFK they notice it. When people on the other team rage and AFK they don't notice it. This of course does make sense since everyone fucking complains about "always having the AFK's/ragers on their team". Everyone can't be above average so at least some of those people must be wrong that they get more AFK's ragers than the other side. But if you legitimately do get more AFK's/ragers you need to look at your behavior and see what is causing it. I meant 9 different multimillion faced dice, 1 for each different player outside of our control player. Also, the bold part is pretty shitty, as I've been pretty civil so far. Not shitty, just likelihood; the proof relies on the derivative of the ratio of two different normal distributions being negative. Not many people are equipped to understand what is going on. Given that you don't seem to understand the LLN as it applies to league with simple binomial distributions I am not going to assume i can make you understand one that relies on the calculus of continuous ones. Anywho; It doesn't matter if its 9 different multimillion faced dice (i mean, its not) since someone is either a rager or not, the result is binary. The fact that a probability = .11012313422 is different than p=.11012313921 doesn't make much of a difference in the overall value. If you're legitimately getting more ragers/afk's/trolls then other people then either the likelihoods are that either A: You're one of them or B: something you're doing is causing it You're being sufficiently vague throwing around the mathematical terms in that first paragraph that I (and I graduated with a minor in math) have no idea what you mean. (I don't really think that discussion should occur, being that it's essentially an argument of whether ELO hell exists or not. I don't care one way or another, but I would like it if you clarified what you meant mathematically.) But you're right about the fact that it has no impact what value p has or how you derive it. It was an aside about discussing ELO hell because Gahlo is essentially saying that ELO hell can exist because you can totally just get unlucky. Specifically if there exists a population of players which are better than their ELO would suggest (say by constantly dodging dropping MMR when that was a thing) then games can get easier as you win and harder as you lose(rather than the other way around) if the probability that you will run into a smurf is decreasing as you increase in MMR. This probability is the ratio of the distributions of "properly placed players" and "smurfs", both of which are normal distributions. In order to actually make an ELO Hell there almost must also be a sufficient density of smurfs because the probability decrease in smurfs has to counteract the difficulty in playing better opponents normally. The probability has to be decrease so the derivative of the ratio of the weighted distributions has to be negative. Going backwards from there shouldn't be that difficult if you knew what was being discussed. This is one of the reasons that Riot stopped having MMR penalties for dodging, because if enough people dodge games consistently it can create that effect. But this doesn't exist as we know it in league. ELO hell aint real. On August 06 2014 09:34 Gahlo wrote:
I'm not having an issue with people in ranked games, to which they said they kept track of how many it was before it got too frustrating to do so. Then again, instead of dealing with actual data you'd probably just dismiss it out of hand. For the record, I was pointing out your misunderstanding for clarity's sake, not that I thought it makes a significant difference. On this case i will absolutely dismiss it out of hand because biased observers cannot generate reliable data. When they play a game they have no way to assess the raginess of the other team except ask the other team. Which A) they're unlikely do do and B) Even if they do they won't get a response that they can compare to their own data and C) even if they do and they get a response that they can compare to their own data [which they can't] they still don't know if they're the problem. The only way to get reliable data is to have an unbiased observer watch the replays from both sides including chat and judge the number of ragers on each side. But nobody has time for that shit. I mean look i've probably watched games from like 10-12 different people who have said "i am totally in ELO hell and am playing just fine" and every time i say "No, its definitely you. You're making lots of mistakes and probably raging" and i go and watch their games and yup, they're making mistakes and throwing and raging and contributing to losses. Sure 40-50 games for 10 people isn't much of a sample size but frankly i don't think anyone has done this type of thing and been like "yea you just totally get screwed in all your games" Good job putting words into my mouth. That's definitely not my definition of ELO hell. Then again, it's probably too plebian for you to understand. When you say that people saying that "one person won't stop your rise" is the gambler's fallacy what do you expect people to think? And that "there won't be an even distribution of trolls/ragers because no one plays that many games"? Note that the gambler's fallacy doesn't actually exist in League because winning more legitimately makes the game harder as does losing make the game easier (so long as your performance also doesn't correlate with wins/losses) The more this line of discussion goes, it becomes more apparent to me that you aren't even reading my posts or just scimming them and assuming you know what I'm saying. Whether you like it or not, it is possible for a player to get a bunch of shit teammates in a decent sized cluster of game and regardless of that, there is never a guarantee in any form or fashion that the tide would reverse.
Mean reversion is a thing. The only way the tide won't "reverse" towards the mean is if that specific player has a different mean. Otherwise as they play more games they will go towards the mean. Over one more game they might not. Over two more games they might not. Over 100 games they almost most definitely will and the only way to say that they will not is quibble with "well there is a half a % chance they do not". This doesn't happen because the prior game indicates that they get less ragers the next game, but because the effect of the initial games on the sample mean goes to zero as the number of games played increases.
|
On August 06 2014 11:06 PrinceXizor wrote:Show nested quote +On August 06 2014 10:57 Goumindong wrote:On August 06 2014 10:51 PrinceXizor wrote:
Look, this is a dumb conversation. especially when one side refuses to acknowledge that there can even be one person stuck at a lower rank than he/she belongs. if its truly random whether or not a player is a rager/afker there are some people who have that rager on the other team disproportionately more than on their team, and some that have it disproportionately on their team compared to the other team. there are always going to be outliers, an argument against outliers is not "well on average it shouldn't matter". Sure there can be outliers but for any "outlier" it significantly more likely they're the problem than they're unlucky. That is to say that outliers may be theoretically possible but good luck finding them thats not how it works mate.
Why not? What is more likely, that you're the .005% or that you're causing the problems?
|
|
|
|
|
|
EPT
