|
On July 05 2013 22:22 highsis wrote:Show nested quote +On July 05 2013 20:23 Hattori_Hanzo wrote:On July 05 2013 13:51 highsis wrote:On July 05 2013 06:15 iky43210 wrote:On July 05 2013 05:12 TheRabidDeer wrote: I am confused, I thought the most recent winrates reported were pretty even across the board? I know what you posted is how it feels, but is it accurate? its not accurate, he is totally bullshitting and cherry picking his data. This is Korea only win rates (international looks even more balanced) http://i.imgur.com/0ciWmTd.png Not "only PL RO 6, finals of WCS (all 6 matches of it), during raining days and only innovations" arbitrary picked data. mine is stat of the highest level leagues from last 30 days or so zerg/ protoss are not doing well at the top level Learn to statistics. Your sample size is bad and you should feel bad. A proper sample size is to include ALL match ups since the last patch. Also, Terran pros are insane. Hello, muthertrucking FLASH and INNOVATION?!!! A fair statistics will remove ALL TOP RACE Pros. That's because one star can bugger all the stats in the world. See Portugal soccer team's Cristian Ronaldo and Russian Chess teams. Foreigners and code A/B level progamers don't represent balance at the highest level which is only relevant. And pros at the very top represent the skillcap and potential that race has, which other pro gamers will eventually catch up. They shouldn't be excluded. if you ever watched BW, top players led to the imrpovement of their respective race. Also about the sample size, you need to consider 'lengths' the data was collected from. ever since HOTS came out there must have been thousands of games that have been played between pros, and the on-screen games are the finest, best skill refined with countless practices and training. If for 3 months one race has been overperforming, that definitely indicate indicate something. In addition, change in game strategy and trend makes it impossible to gain a data collectively representing the 'overal' balance because it shifts from time to time, so a compromise must be made as well as in defining the range of the highest level group. If one player wins 9 games out of 10 you could still say that he was just lucky because there are only 10 games sample size, but it's more feasible to believe he is actually playing better than other players at the moment. Larger sample size will provide more accurate answers, but even with smaller sample size you can get the flow of the balance. And last for the love of mankind stop calling me ignorant or yelling me to learn. Some comments are unnecessarily aggressive, much more so than your average TL comments.
I build statistical models for a living, I think you should show some humility. Basics of statistics, identify and remove anomalies FIRST even before any samples are collected. You won't learn this in statics 101 because most professors do not depend on the accuracy of their models for their rent money.
Where I come from, 300 sample size is THE MINIMUM before any analysis can begin.
All any pro needs to throw win ratios out of whack is an unfair advantage repeated over an extended period.
Which is why using average statistics without consideration of such factors will lead to bad decisions e.g. Queen buffs that massively affected Terran, eliminating an entire branch of harass because Zergs weren't using queens for anything but larvae duty.
|
On July 06 2013 01:48 Hattori_Hanzo wrote:Show nested quote +On July 05 2013 22:22 highsis wrote:On July 05 2013 20:23 Hattori_Hanzo wrote:On July 05 2013 13:51 highsis wrote:On July 05 2013 06:15 iky43210 wrote:On July 05 2013 05:12 TheRabidDeer wrote: I am confused, I thought the most recent winrates reported were pretty even across the board? I know what you posted is how it feels, but is it accurate? its not accurate, he is totally bullshitting and cherry picking his data. This is Korea only win rates (international looks even more balanced) http://i.imgur.com/0ciWmTd.png Not "only PL RO 6, finals of WCS (all 6 matches of it), during raining days and only innovations" arbitrary picked data. mine is stat of the highest level leagues from last 30 days or so zerg/ protoss are not doing well at the top level Learn to statistics. Your sample size is bad and you should feel bad. A proper sample size is to include ALL match ups since the last patch. Also, Terran pros are insane. Hello, muthertrucking FLASH and INNOVATION?!!! A fair statistics will remove ALL TOP RACE Pros. That's because one star can bugger all the stats in the world. See Portugal soccer team's Cristian Ronaldo and Russian Chess teams. Foreigners and code A/B level progamers don't represent balance at the highest level which is only relevant. And pros at the very top represent the skillcap and potential that race has, which other pro gamers will eventually catch up. They shouldn't be excluded. if you ever watched BW, top players led to the imrpovement of their respective race. Also about the sample size, you need to consider 'lengths' the data was collected from. ever since HOTS came out there must have been thousands of games that have been played between pros, and the on-screen games are the finest, best skill refined with countless practices and training. If for 3 months one race has been overperforming, that definitely indicate indicate something. In addition, change in game strategy and trend makes it impossible to gain a data collectively representing the 'overal' balance because it shifts from time to time, so a compromise must be made as well as in defining the range of the highest level group. If one player wins 9 games out of 10 you could still say that he was just lucky because there are only 10 games sample size, but it's more feasible to believe he is actually playing better than other players at the moment. Larger sample size will provide more accurate answers, but even with smaller sample size you can get the flow of the balance. And last for the love of mankind stop calling me ignorant or yelling me to learn. Some comments are unnecessarily aggressive, much more so than your average TL comments. I build statistical models for a living, I think you should show some humility. Basics of statistics, identify and remove anomalies FIRST even before any samples are collected. You won't learn this in statics 101 because most professors do not depend on the accuracy of their models for their rent money. Where I come from, 300 sample size is THE MINIMUM before any analysis can begin. Quite frankly if highsis would have followed statistics 101 he wouldn't have made those claims in the first place, even in statistics 101 you learn something regarding significance of data.
|
On July 05 2013 22:22 highsis wrote:Show nested quote +On July 05 2013 20:23 Hattori_Hanzo wrote:On July 05 2013 13:51 highsis wrote:On July 05 2013 06:15 iky43210 wrote:On July 05 2013 05:12 TheRabidDeer wrote: I am confused, I thought the most recent winrates reported were pretty even across the board? I know what you posted is how it feels, but is it accurate? its not accurate, he is totally bullshitting and cherry picking his data. This is Korea only win rates (international looks even more balanced) http://i.imgur.com/0ciWmTd.png Not "only PL RO 6, finals of WCS (all 6 matches of it), during raining days and only innovations" arbitrary picked data. mine is stat of the highest level leagues from last 30 days or so zerg/ protoss are not doing well at the top level Learn to statistics. Your sample size is bad and you should feel bad. A proper sample size is to include ALL match ups since the last patch. Also, Terran pros are insane. Hello, muthertrucking FLASH and INNOVATION?!!! A fair statistics will remove ALL TOP RACE Pros. That's because one star can bugger all the stats in the world. See Portugal soccer team's Cristian Ronaldo and Russian Chess teams. Foreigners and code A/B level progamers don't represent balance at the highest level which is only relevant. And pros at the very top represent the skillcap and potential that race has, which other pro gamers will eventually catch up. They shouldn't be excluded. if you ever watched BW, top players led to the imrpovement of their respective race. Also about the sample size, you need to consider 'lengths' the data was collected from. ever since HOTS came out there must have been thousands of games that have been played between pros, and the on-screen games are the finest, best skill refined with countless practices and training. If for 3 months one race has been overperforming, that definitely indicate indicate something. In addition, change in game strategy and trend makes it impossible to gain a data collectively representing the 'overal' balance because it shifts from time to time, so a compromise must be made as well as in defining the range of the highest level group. If one player wins 9 games out of 10 you could still say that he was just lucky because there are only 10 games sample size, but it's more feasible to believe he is actually playing better than other players at the moment. Larger sample size will provide more accurate answers, but even with smaller sample size you can get the flow of the balance. And last for the love of mankind stop calling me ignorant or yelling me to learn. Some comments are unnecessarily aggressive, much more so than your average TL comments.
That is not necessarily true. You need to describe your sampling method as well. A large sample and its data may not mean anything if there is bias involved. Hence, a random sample (or a group of random samples) is much more important than a large one.
EDIT: Furthermore, I do not understand how you can ignore foreigners and Code A/B people when they can, and often do, contribute to the highest level of play
|
On July 06 2013 02:49 jkim91 wrote:Show nested quote +On July 05 2013 22:22 highsis wrote:On July 05 2013 20:23 Hattori_Hanzo wrote:On July 05 2013 13:51 highsis wrote:On July 05 2013 06:15 iky43210 wrote:On July 05 2013 05:12 TheRabidDeer wrote: I am confused, I thought the most recent winrates reported were pretty even across the board? I know what you posted is how it feels, but is it accurate? its not accurate, he is totally bullshitting and cherry picking his data. This is Korea only win rates (international looks even more balanced) http://i.imgur.com/0ciWmTd.png Not "only PL RO 6, finals of WCS (all 6 matches of it), during raining days and only innovations" arbitrary picked data. mine is stat of the highest level leagues from last 30 days or so zerg/ protoss are not doing well at the top level Learn to statistics. Your sample size is bad and you should feel bad. A proper sample size is to include ALL match ups since the last patch. Also, Terran pros are insane. Hello, muthertrucking FLASH and INNOVATION?!!! A fair statistics will remove ALL TOP RACE Pros. That's because one star can bugger all the stats in the world. See Portugal soccer team's Cristian Ronaldo and Russian Chess teams. Foreigners and code A/B level progamers don't represent balance at the highest level which is only relevant. And pros at the very top represent the skillcap and potential that race has, which other pro gamers will eventually catch up. They shouldn't be excluded. if you ever watched BW, top players led to the imrpovement of their respective race. Also about the sample size, you need to consider 'lengths' the data was collected from. ever since HOTS came out there must have been thousands of games that have been played between pros, and the on-screen games are the finest, best skill refined with countless practices and training. If for 3 months one race has been overperforming, that definitely indicate indicate something. In addition, change in game strategy and trend makes it impossible to gain a data collectively representing the 'overal' balance because it shifts from time to time, so a compromise must be made as well as in defining the range of the highest level group. If one player wins 9 games out of 10 you could still say that he was just lucky because there are only 10 games sample size, but it's more feasible to believe he is actually playing better than other players at the moment. Larger sample size will provide more accurate answers, but even with smaller sample size you can get the flow of the balance. And last for the love of mankind stop calling me ignorant or yelling me to learn. Some comments are unnecessarily aggressive, much more so than your average TL comments. That is not necessarily true. You need to describe your sampling method as well. A large sample and its data may not mean anything if there is bias involved. Hence, a random sample (or a group of random samples) is much more important than a large one.
I think the point was that a larger same will always be more accurate than a small sample. You can't just assume the larger sample has bias, and hence the random smaller sample is better. If you have a larger random sample (with no bias), that is better than a small one of the same.. There is no case in which more data is a bad thing as long as it's equally accurate/useable. You can't just assume the larger data has bias and the smaller data doesn't. I think the point you were trying to make is that less data that unbiased is better than more data that is biased. That is usually true but that's not the point the other person was arguing. If both sets of data are equal on terms of accuracy, the larger set is better and more useful/accurate.
|
On July 05 2013 22:22 highsis wrote:Show nested quote +On July 05 2013 20:23 Hattori_Hanzo wrote:On July 05 2013 13:51 highsis wrote:On July 05 2013 06:15 iky43210 wrote:On July 05 2013 05:12 TheRabidDeer wrote: I am confused, I thought the most recent winrates reported were pretty even across the board? I know what you posted is how it feels, but is it accurate? its not accurate, he is totally bullshitting and cherry picking his data. This is Korea only win rates (international looks even more balanced) http://i.imgur.com/0ciWmTd.png Not "only PL RO 6, finals of WCS (all 6 matches of it), during raining days and only innovations" arbitrary picked data. mine is stat of the highest level leagues from last 30 days or so zerg/ protoss are not doing well at the top level Learn to statistics. Your sample size is bad and you should feel bad. A proper sample size is to include ALL match ups since the last patch. Also, Terran pros are insane. Hello, muthertrucking FLASH and INNOVATION?!!! A fair statistics will remove ALL TOP RACE Pros. That's because one star can bugger all the stats in the world. See Portugal soccer team's Cristian Ronaldo and Russian Chess teams. Foreigners and code A/B level progamers don't represent balance at the highest level which is only relevant. And pros at the very top represent the skillcap and potential that race has, which other pro gamers will eventually catch up. They shouldn't be excluded. if you ever watched BW, top players led to the imrpovement of their respective race. Also about the sample size, you need to consider 'lengths' the data was collected from. ever since HOTS came out there must have been thousands of games that have been played between pros, and the on-screen games are the finest, best skill refined with countless practices and training. If for 3 months one race has been overperforming, that definitely indicate indicate something. In addition, change in game strategy and trend makes it impossible to gain a data collectively representing the 'overal' balance because it shifts from time to time, so a compromise must be made as well as in defining the range of the highest level group. If one player wins 9 games out of 10 you could still say that he was just lucky because there are only 10 games sample size, but it's more feasible to believe he is actually playing better than other players at the moment. Larger sample size will provide more accurate answers, but even with smaller sample size you can get the flow of the balance. And last for the love of mankind stop calling me ignorant or yelling me to learn. Some comments are unnecessarily aggressive, much more so than your average TL comments.
You should take the hint if everyone is criticizing your claims and there's no-one rushing to your aid. I'll also add to criticism.
You said that you cannot take foreigners, Code A and Code B into account. But that doesn't correspond to your statistics which includes proleague and Code S. First of all, Code A feeds into Code S, and someone can play a Code S group, lose, but win the Code A match to return straight to Code S. Does that mean the player was significantly better in the Code S games, than in the Code A match. No.
Secondly, the proleague includes B-teamers (hello Bear) and foreigners (EG-Liquid guys), yet you included it into your sample, ignoring qualms people had with the format and maps used.
A small sample can be indicative of trends, but it might not be. For example, it's a true that you can exit poll a small proportion of the electorate to accurately predict the results of the elections that include significantly more people. Yet, even these can be false. On the other hand, it's not accurate to say that 3 T wins agains 2 zerg losses is indicative of anything except for random chance as the number of games is so small, it's extremely easy to add data and have the result be skewed in the opposite direction.
|
On July 06 2013 03:22 onPHYRE wrote:Show nested quote +On July 06 2013 02:49 jkim91 wrote:On July 05 2013 22:22 highsis wrote:On July 05 2013 20:23 Hattori_Hanzo wrote:On July 05 2013 13:51 highsis wrote:On July 05 2013 06:15 iky43210 wrote:On July 05 2013 05:12 TheRabidDeer wrote: I am confused, I thought the most recent winrates reported were pretty even across the board? I know what you posted is how it feels, but is it accurate? its not accurate, he is totally bullshitting and cherry picking his data. This is Korea only win rates (international looks even more balanced) http://i.imgur.com/0ciWmTd.png Not "only PL RO 6, finals of WCS (all 6 matches of it), during raining days and only innovations" arbitrary picked data. mine is stat of the highest level leagues from last 30 days or so zerg/ protoss are not doing well at the top level Learn to statistics. Your sample size is bad and you should feel bad. A proper sample size is to include ALL match ups since the last patch. Also, Terran pros are insane. Hello, muthertrucking FLASH and INNOVATION?!!! A fair statistics will remove ALL TOP RACE Pros. That's because one star can bugger all the stats in the world. See Portugal soccer team's Cristian Ronaldo and Russian Chess teams. Foreigners and code A/B level progamers don't represent balance at the highest level which is only relevant. And pros at the very top represent the skillcap and potential that race has, which other pro gamers will eventually catch up. They shouldn't be excluded. if you ever watched BW, top players led to the imrpovement of their respective race. Also about the sample size, you need to consider 'lengths' the data was collected from. ever since HOTS came out there must have been thousands of games that have been played between pros, and the on-screen games are the finest, best skill refined with countless practices and training. If for 3 months one race has been overperforming, that definitely indicate indicate something. In addition, change in game strategy and trend makes it impossible to gain a data collectively representing the 'overal' balance because it shifts from time to time, so a compromise must be made as well as in defining the range of the highest level group. If one player wins 9 games out of 10 you could still say that he was just lucky because there are only 10 games sample size, but it's more feasible to believe he is actually playing better than other players at the moment. Larger sample size will provide more accurate answers, but even with smaller sample size you can get the flow of the balance. And last for the love of mankind stop calling me ignorant or yelling me to learn. Some comments are unnecessarily aggressive, much more so than your average TL comments. That is not necessarily true. You need to describe your sampling method as well. A large sample and its data may not mean anything if there is bias involved. Hence, a random sample (or a group of random samples) is much more important than a large one. I think the point was that a larger same will always be more accurate than a small sample. You can't just assume the larger sample has bias, and hence the random smaller sample is better. If you have a larger random sample (with no bias), that is better than a small one of the same.. There is no case in which more data is a bad thing as long as it's equally accurate/useable. You can't just assume the larger data has bias and the smaller data doesn't. I think the point you were trying to make is that less data that unbiased is better than more data that is biased. That is usually true but that's not the point the other person was arguing. If both sets of data are equal on terms of accuracy, the larger set is better and more useful/accurate.
What the hell are you talking about? My point was that you need to state your sampling method if any conclusion you draw from the data is to be considered accurate. I'm not assuming that large data is always biased compared to a small one. That is just an example.
|
|
|
|