|
On November 11 2012 02:57 Toxi78 wrote:Show nested quote +On November 11 2012 02:41 TheBB wrote:On November 11 2012 02:40 DOUDOU wrote:On November 11 2012 02:37 TheBB wrote:On November 11 2012 02:32 Toxi78 wrote:
then there is something totally moronic about the way you actually get to these percentages? polt and stephano being in the same bracket, having 40% for stephano to win and 7% for polt to win makes little to no sense?
just the fact that you throw out these statistics like it means anything is totally hilarious.
why not compare the last 5 bo3s of the players (regardless of who they played) so you get their shape, then make a totally ridiculous stat in which lets say axslav won his last 5 matches and stephano lost his last 5 and have a ranking that adds up to 100% and doesnt look mathematically flawed yet makes no sense?
the fact that mathematically your stats make sense doesn't mean that the method is not flawed.
your statistics are bullshit, there is so much statistical bias in them that they are just worth laughing at. The method is described in detail here: http://www.teamliquid.net/blogs/viewblog.php?id=380850Feel free to make specific and constructive criticism there instead of polluting this thread. I never claimed they mean anything. It's up to everyone to interpret them as they please. i'm no statistician but is 0% even suppose to be possible? Ah, no, they're just estimates from a large amount of samples (50k). Qxc has a very small probability of winning. It's not zero but it's so small that it might as well be. (Poor guy.) now run the script to see what his chances were to all-kill IM with beating MVP. yet it happend. because sc2 is nothing like chess and ELO rankings don't make a lot of sense in it. sure you did a good job with all this programming, as a statistician i reckon it's a bit of work (though far from being very complicated for anyone that has a bit of knowledge, but you fail to see that the problems come from the method and not the mathematical background. when programming something like this, the question is not what you develop the most in your thread aka the mathematical proof (anyone with some experience in R or python or w/e can do that), but the way you get the probability of player A beating player B. as you say yourself in the thread "if you accept that my Elo model is a perfect predictor". This is hilarious. Making statistics and using the biggest authority argument ever : "elo model is a perfect predictor". I'm just trying to open your eyes, great things could be done (and i'd be very interested to work in this or at least give it a few thoughts). All you need is not a huge perfect ranking of all active players to make such stats, it's the most accurate way to describe a player's chances to beat another player. we both know you and me that starcraft 2 is a not a transitive game so basically basing your whole analysis on ELO which is a ranking that pretty much compares the wins and loses of a player vs everyone against the wins and loses of the other player doesn't make a lot of sense to me. we can further develop by PMs if you want. i'm sorry i was rude at first. ps: why not use R instead of python, i find python terrible. could you also please detail the probability of player A to beat player B based on their ELOs, i'm interested anyways.
And statistically improbable things happen. Not sure what your qxc example was suppose to show. Are you saying there is a mathematical model which would have predicted that qxc would AK IM?
Of course, ELO has its flaws. Some does any statistical model when so many variables are involved. But to say ELO only applies to chess and not SC2 is just strange. Since the ladder is basically a modified ELO.
|
On November 11 2012 02:57 Toxi78 wrote:Show nested quote +On November 11 2012 02:41 TheBB wrote:On November 11 2012 02:40 DOUDOU wrote:On November 11 2012 02:37 TheBB wrote:On November 11 2012 02:32 Toxi78 wrote:
then there is something totally moronic about the way you actually get to these percentages? polt and stephano being in the same bracket, having 40% for stephano to win and 7% for polt to win makes little to no sense?
just the fact that you throw out these statistics like it means anything is totally hilarious.
why not compare the last 5 bo3s of the players (regardless of who they played) so you get their shape, then make a totally ridiculous stat in which lets say axslav won his last 5 matches and stephano lost his last 5 and have a ranking that adds up to 100% and doesnt look mathematically flawed yet makes no sense?
the fact that mathematically your stats make sense doesn't mean that the method is not flawed.
your statistics are bullshit, there is so much statistical bias in them that they are just worth laughing at. The method is described in detail here: http://www.teamliquid.net/blogs/viewblog.php?id=380850Feel free to make specific and constructive criticism there instead of polluting this thread. I never claimed they mean anything. It's up to everyone to interpret them as they please. i'm no statistician but is 0% even suppose to be possible? Ah, no, they're just estimates from a large amount of samples (50k). Qxc has a very small probability of winning. It's not zero but it's so small that it might as well be. (Poor guy.) now run the script to see what his chances were to all-kill IM with beating MVP. yet it happend. because sc2 is nothing like chess and ELO rankings don't make a lot of sense in it. sure you did a good job with all this programming, as a statistician i reckon it's a bit of work (though far from being very complicated for anyone that has a bit of knowledge, but you fail to see that the problems come from the method and not the mathematical background. when programming something like this, the question is not what you develop the most in your thread aka the mathematical proof (anyone with some experience in R or python or w/e can do that), but the way you get the probability of player A beating player B. as you say yourself in the thread "if you accept that my Elo model is a perfect predictor". This is hilarious. Making statistics and using the biggest authority argument ever : "elo model is a perfect predictor". I'm just trying to open your eyes, great things could be done (and i'd be very interested to work in this or at least give it a few thoughts). All you need is not a huge perfect ranking of all active players to make such stats, it's the most accurate way to describe a player's chances to beat another player. we both know you and me that starcraft 2 is a not a transitive game so basically basing your whole analysis on ELO which is a ranking that pretty much compares the wins and loses of a player vs everyone against the wins and loses of the other player doesn't make a lot of sense to me. we can further develop by PMs if you want. i'm sorry i was rude at first. ps: why not use R instead of python, i find python terrible. could you also please detail the probability of player A to beat player B based on their ELOs, i'm interested anyways.
Yeah, I would be happy to get suggestions, but you don't offer very specific ones. Yet, anyway.
As a statistician you surely know that improbable events happen all the time. (See that famous Feynman quote that I'm sure you've heard.)
The blog post was aimed at the hordes of people asking "where do you get these numbers?" — people with high school maths knowledge at best. Not working mathematicians. My hypothetical assumption that Elo is a perfect predictor (and I know very well it is not) was mean to clarify readers that that is where the potential problems lie, and that everything after is genuine (up to a point, such as independence between games).
I know Elo is hardly perfect, but it's what we have at the moment, for better or worse. I'm sorry I couldn't offer you something mindblowing. This is a three weeks old project based on my free time only. People (mostly) seem to enjoy it for what it is, and if I or someone else can make it into something better, then all the more power to us.
I don't know R and Python is awesome. :D
(Apology accepted. Suggest we keep this to PMs or the blog post from now on.)
|
United Kingdom50293 Posts
Ganzi welcomes all drone donations from ret.
|
On November 11 2012 02:57 Toxi78 wrote:Show nested quote +On November 11 2012 02:41 TheBB wrote:On November 11 2012 02:40 DOUDOU wrote:On November 11 2012 02:37 TheBB wrote:On November 11 2012 02:32 Toxi78 wrote:
then there is something totally moronic about the way you actually get to these percentages? polt and stephano being in the same bracket, having 40% for stephano to win and 7% for polt to win makes little to no sense?
just the fact that you throw out these statistics like it means anything is totally hilarious.
why not compare the last 5 bo3s of the players (regardless of who they played) so you get their shape, then make a totally ridiculous stat in which lets say axslav won his last 5 matches and stephano lost his last 5 and have a ranking that adds up to 100% and doesnt look mathematically flawed yet makes no sense?
the fact that mathematically your stats make sense doesn't mean that the method is not flawed.
your statistics are bullshit, there is so much statistical bias in them that they are just worth laughing at. The method is described in detail here: http://www.teamliquid.net/blogs/viewblog.php?id=380850Feel free to make specific and constructive criticism there instead of polluting this thread. I never claimed they mean anything. It's up to everyone to interpret them as they please. i'm no statistician but is 0% even suppose to be possible? Ah, no, they're just estimates from a large amount of samples (50k). Qxc has a very small probability of winning. It's not zero but it's so small that it might as well be. (Poor guy.) now run the script to see what his chances were to all-kill IM with beating MVP. yet it happend. because sc2 is nothing like chess and ELO rankings don't make a lot of sense in it. sure you did a good job with all this programming, as a statistician i reckon it's a bit of work (though far from being very complicated for anyone that has a bit of knowledge, but you fail to see that the problems come from the method and not the mathematical background. when programming something like this, the question is not what you develop the most in your thread aka the mathematical proof (anyone with some experience in R or python or w/e can do that), but the way you get the probability of player A beating player B. as you say yourself in the thread "if you accept that my Elo model is a perfect predictor". This is hilarious. Making statistics and using the biggest authority argument ever : "elo model is a perfect predictor". I'm just trying to open your eyes, great things could be done (and i'd be very interested to work in this or at least give it a few thoughts). All you need is not a huge perfect ranking of all active players to make such stats, it's the most accurate way to describe a player's chances to beat another player. we both know you and me that starcraft 2 is a not a transitive game so basically basing your whole analysis on ELO which is a ranking that pretty much compares the wins and loses of a player vs everyone against the wins and loses of the other player doesn't make a lot of sense to me. we can further develop by PMs if you want. i'm sorry i was rude at first. ps: why not use R instead of python, i find python terrible. could you also please detail the probability of player A to beat player B based on their ELOs, i'm interested anyways.
I am not sure whether you are desperately trying to pose as a competent person in order to give authority to your argument or if you are the competent person who is absolute least capable of expressing himself that I have ever met in my life. Except I am quite sure it's the first. 
Chill, nobody cares about the little skirmishes in LR threads. Live and let live. 
|
|
|
|
|
|
|
Blunders...
Blunders everywhere!!
|
|
|
Ret just went full foreigner
|
United States97276 Posts
|
United Kingdom14464 Posts
Sloppy game, nice control from Ganzi though.
|
On November 11 2012 03:07 PhoenixVoid wrote:
What is Ret doing...
Going full foreigner mode
|
United Kingdom50293 Posts
Sometimes I wonder how ret even wins...
|
zerg players are so fucking bad at unit splitting
|
On November 11 2012 03:07 Mackus wrote:
Ret just went full foreigner
He does this all the time. constantly blows huge leads
|
|
|
Is it just me or does Ret always look like a platinum level player?
|
Wow. Ret plays like he has completely given up on trying at this game.
|
Poor games from Ret, really poor. Hope he can pick it up in the lower bracket!
|
|
|
|
|
|
EPT
