|
On September 21 2013 12:20 EatThePath wrote:Show nested quote +On September 20 2013 05:45 Die4Ever wrote:On September 20 2013 05:29 KillerDucky wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. He did a run without Aligulac, and that case for example Revival has exactly 50/50 chance to win/lose. But in the monte-carlo results is says Revival has 49.85/50.15 chance, off in the first decimal place. + Show Spoiler [example] +
Revival, 243912/300000, started with 2900 WCS points, 81.304%
Revival starts in the round of 32 in America Premier facing Polt, Sage, HyuN
Revival loses this match 49.85% of the time, which changes Revival's chances to 71.56%.
Revival wins this match 50.15% of the time, which changes Revival's chances to 90.99%.
My concern isn't so much about accuracy, it's more just I don't want 0.03% to be displayed as 0%. I should read more before I ask but I'm lazy (sorry) and you might just have the answer handy anyway, but is there a confidence interval with your %chance results and an error range?
There is not. I don't know how to calculate that overall. Per match it seems to be about 0.3% though.
|
On September 21 2013 12:49 Die4Ever wrote:Show nested quote +On September 21 2013 12:20 EatThePath wrote:On September 20 2013 05:45 Die4Ever wrote:On September 20 2013 05:29 KillerDucky wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. He did a run without Aligulac, and that case for example Revival has exactly 50/50 chance to win/lose. But in the monte-carlo results is says Revival has 49.85/50.15 chance, off in the first decimal place. + Show Spoiler [example] +
Revival, 243912/300000, started with 2900 WCS points, 81.304%
Revival starts in the round of 32 in America Premier facing Polt, Sage, HyuN
Revival loses this match 49.85% of the time, which changes Revival's chances to 71.56%.
Revival wins this match 50.15% of the time, which changes Revival's chances to 90.99%.
My concern isn't so much about accuracy, it's more just I don't want 0.03% to be displayed as 0%. I should read more before I ask but I'm lazy (sorry) and you might just have the answer handy anyway, but is there a confidence interval with your %chance results and an error range? There is not. I don't know how to calculate that overall. Per match it seems to be about 0.3% though.
Okay. Thanks 
|
On September 20 2013 04:19 EatThePath wrote:Show nested quote +On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo.
The input data is not good enough. If you have only a handful of samples from the past, you cannot justify high-precision probability.
|
On September 22 2013 20:32 [F_]aths wrote:Show nested quote +On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. The input data is not good enough. If you have only a handful of samples from the past, you cannot justify high-precision probability.
Ok I get it, you want less precise numbers. Thank you for the feedback.
|
On September 23 2013 03:15 Die4Ever wrote:Show nested quote +On September 22 2013 20:32 [F_]aths wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. The input data is not good enough. If you have only a handful of samples from the past, you cannot justify high-precision probability. Ok I get it, you want less precise numbers. Thank you for the feedback.
I think one digit is fine, it is important for the single digit chances and the near 100% chances.
|
On September 21 2013 12:49 Die4Ever wrote:Show nested quote +On September 21 2013 12:20 EatThePath wrote:On September 20 2013 05:45 Die4Ever wrote:On September 20 2013 05:29 KillerDucky wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. He did a run without Aligulac, and that case for example Revival has exactly 50/50 chance to win/lose. But in the monte-carlo results is says Revival has 49.85/50.15 chance, off in the first decimal place. + Show Spoiler [example] +
Revival, 243912/300000, started with 2900 WCS points, 81.304%
Revival starts in the round of 32 in America Premier facing Polt, Sage, HyuN
Revival loses this match 49.85% of the time, which changes Revival's chances to 71.56%.
Revival wins this match 50.15% of the time, which changes Revival's chances to 90.99%.
My concern isn't so much about accuracy, it's more just I don't want 0.03% to be displayed as 0%. I should read more before I ask but I'm lazy (sorry) and you might just have the answer handy anyway, but is there a confidence interval with your %chance results and an error range? There is not. I don't know how to calculate that overall. Per match it seems to be about 0.3% though.
I would say 2 * sqrt(success chance - success chance ^2 ) / sqrt (number of runs) should give approximatly a 2 sigma confidence interval, which for a gaussian (which it isn't but it might work as approximation) would be a 95% confidence interval.
So if 30000 runs give a 50% sucess rate, the accuracy would be +/- 0.3% chance as you said. For all other chance it should be lower.
|
----NaNiwa gets 32nd place in IEM
This happens 43.75% of the time. When it does, it changes NaNiwa's chances to 3.40%.
Um, is this not theoretically impossible as Naniwa is seeded directly into the round of 16?
|
On September 23 2013 21:54 Ponchey wrote:Show nested quote +----NaNiwa gets 32nd place in IEM
This happens 43.75% of the time. When it does, it changes NaNiwa's chances to 3.40%. Um, is this not theoretically impossible as Naniwa is seeded directly into the round of 16?
sorry, I thought that the qualified players started in ro32, will be fixed in the next update
|
Awesome. And thanks for this, it's great!
|
On September 22 2013 20:32 [F_]aths wrote:Show nested quote +On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. The input data is not good enough. If you have only a handful of samples from the past, you cannot justify high-precision probability.
I may be wrong, but in monte carlo you just take the bayesian network as an assumption, and the number of trials is what determines your precision? Of course realistically it all depends on the accuracy of your underlying probabilities.
On September 23 2013 05:51 Sandermatt wrote:Show nested quote +On September 21 2013 12:49 Die4Ever wrote:On September 21 2013 12:20 EatThePath wrote:On September 20 2013 05:45 Die4Ever wrote:On September 20 2013 05:29 KillerDucky wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. He did a run without Aligulac, and that case for example Revival has exactly 50/50 chance to win/lose. But in the monte-carlo results is says Revival has 49.85/50.15 chance, off in the first decimal place. + Show Spoiler [example] +
Revival, 243912/300000, started with 2900 WCS points, 81.304%
Revival starts in the round of 32 in America Premier facing Polt, Sage, HyuN
Revival loses this match 49.85% of the time, which changes Revival's chances to 71.56%.
Revival wins this match 50.15% of the time, which changes Revival's chances to 90.99%.
My concern isn't so much about accuracy, it's more just I don't want 0.03% to be displayed as 0%. I should read more before I ask but I'm lazy (sorry) and you might just have the answer handy anyway, but is there a confidence interval with your %chance results and an error range? There is not. I don't know how to calculate that overall. Per match it seems to be about 0.3% though. I would say 2 * sqrt(success chance - success chance ^2 ) / sqrt (number of runs) should give approximatly a 2 sigma confidence interval, which for a gaussian (which it isn't but it might work as approximation) would be a 95% confidence interval. So if 30000 runs give a 50% sucess rate, the accuracy would be +/- 0.3% chance as you said. For all other chance it should be lower.
Hmm, okay. Thanks.
|
Technically speaking, he could also make slight modifications to the wording which would allow for whatever precision he wants to be justified.
|
On September 24 2013 03:18 EatThePath wrote:Show nested quote +On September 22 2013 20:32 [F_]aths wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. The input data is not good enough. If you have only a handful of samples from the past, you cannot justify high-precision probability. I may be wrong, but in monte carlo you just take the bayesian network as an assumption, and the number of trials is what determines your precision? Of course realistically it all depends on the accuracy of your underlying probabilities. Show nested quote +On September 23 2013 05:51 Sandermatt wrote:On September 21 2013 12:49 Die4Ever wrote:On September 21 2013 12:20 EatThePath wrote:On September 20 2013 05:45 Die4Ever wrote:On September 20 2013 05:29 KillerDucky wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:
[quote]The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. He did a run without Aligulac, and that case for example Revival has exactly 50/50 chance to win/lose. But in the monte-carlo results is says Revival has 49.85/50.15 chance, off in the first decimal place. + Show Spoiler [example] +
Revival, 243912/300000, started with 2900 WCS points, 81.304%
Revival starts in the round of 32 in America Premier facing Polt, Sage, HyuN
Revival loses this match 49.85% of the time, which changes Revival's chances to 71.56%.
Revival wins this match 50.15% of the time, which changes Revival's chances to 90.99%.
My concern isn't so much about accuracy, it's more just I don't want 0.03% to be displayed as 0%. I should read more before I ask but I'm lazy (sorry) and you might just have the answer handy anyway, but is there a confidence interval with your %chance results and an error range? There is not. I don't know how to calculate that overall. Per match it seems to be about 0.3% though. I would say 2 * sqrt(success chance - success chance ^2 ) / sqrt (number of runs) should give approximatly a 2 sigma confidence interval, which for a gaussian (which it isn't but it might work as approximation) would be a 95% confidence interval. So if 30000 runs give a 50% sucess rate, the accuracy would be +/- 0.3% chance as you said. For all other chance it should be lower. Hmm, okay. Thanks.
I just realised my answer is confusing. By lower I mean the deviation is lower not the accuracy. So if somebody has a 1% chance to qualify it is like +/- 0.01%.
|
On September 24 2013 05:56 Sandermatt wrote:Show nested quote +On September 24 2013 03:18 EatThePath wrote:On September 22 2013 20:32 [F_]aths wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:On September 18 2013 00:25 [F_]aths wrote:On September 13 2013 20:50 Die4Ever wrote:
I've been working on a program that calculates each players' chances of going to Blizzcon. It works by running hundreds of thousands of simulations of the tournament brackets using Monte Carlo method(wikipedia it) with the help of Aligulac ratings. Not only does it give % chances, but it also lists events that help or hurt that player's chances in the details section.
----MMA Acer gets 16th place in Season 3 Finals
This happens 10.1397% of the time. When it does, it changes his chances to 74.7506%. The number of past events does not justify a probability calculation with four decimal places. You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. The input data is not good enough. If you have only a handful of samples from the past, you cannot justify high-precision probability. I may be wrong, but in monte carlo you just take the bayesian network as an assumption, and the number of trials is what determines your precision? Of course realistically it all depends on the accuracy of your underlying probabilities. On September 23 2013 05:51 Sandermatt wrote:On September 21 2013 12:49 Die4Ever wrote:On September 21 2013 12:20 EatThePath wrote:On September 20 2013 05:45 Die4Ever wrote:On September 20 2013 05:29 KillerDucky wrote:On September 20 2013 04:19 EatThePath wrote:On September 19 2013 16:22 [F_]aths wrote:On September 18 2013 01:17 Die4Ever wrote:
[quote]
You're right, maybe I'll change to 1 or 2. Even that is massively excessive, but better than four decimal places. Is it? He's running a lot of monte carlo. He did a run without Aligulac, and that case for example Revival has exactly 50/50 chance to win/lose. But in the monte-carlo results is says Revival has 49.85/50.15 chance, off in the first decimal place. + Show Spoiler [example] +
Revival, 243912/300000, started with 2900 WCS points, 81.304%
Revival starts in the round of 32 in America Premier facing Polt, Sage, HyuN
Revival loses this match 49.85% of the time, which changes Revival's chances to 71.56%.
Revival wins this match 50.15% of the time, which changes Revival's chances to 90.99%.
My concern isn't so much about accuracy, it's more just I don't want 0.03% to be displayed as 0%. I should read more before I ask but I'm lazy (sorry) and you might just have the answer handy anyway, but is there a confidence interval with your %chance results and an error range? There is not. I don't know how to calculate that overall. Per match it seems to be about 0.3% though. I would say 2 * sqrt(success chance - success chance ^2 ) / sqrt (number of runs) should give approximatly a 2 sigma confidence interval, which for a gaussian (which it isn't but it might work as approximation) would be a 95% confidence interval. So if 30000 runs give a 50% sucess rate, the accuracy would be +/- 0.3% chance as you said. For all other chance it should be lower. Hmm, okay. Thanks. I just realised my answer is confusing. By lower I mean the deviation is lower not the accuracy. So if somebody has a 1% chance to qualify it is like +/- 0.01%.
Yeah I get you. That makes sense actually if the system generates such a high or low probability it is not going to change much due to uncertainty, with 50% being the most uncertain answer.
|
Just updated with GSL Group A, WCS AM Group G, and also IEM players starting in ro16.
|
Tomorrow WCS EU Group A with Happy, Targa, Welmu, and Stardust.
Happy, Targa, Welmu, and Stardust all need to advance to have a realistic chance.
WCS AM Group A with Polt, Hyun, Revival, and Sage.
Hyun and Sage need to advance to have a realistic chance. Revival has a 77.8% chance now, goes down to 71.7% if he loses and up to 89.9% if he wins. Polt is safe.
WCS KR Group B with Rain, Keen, sOs, and Trap.
Rain, Keen, and Trap need to advance to have a realistic chance. sOs is pretty safe at 99.471%.
|
Oz, 23185/300000, started with 1100 WCS points, 7.72833%- Hide Spoiler [IF Game Changers] -
Oz starts in the round of 16 in America Premier facing DeMusliM, Apocalypse, Suppy
Oz loses this match 53.25% of the time, which changes Oz's chances to 0.00%.
Oz wins this match 46.75% of the time, which changes Oz's chances to 16.53%.
That was the Ro32 Group F (which you claim has been included), and Oz got through.
|
Could you share some more detailed results for NaNiwa with us? We can see, that if he manages to win IEM, he is almost guaranteed to advance, and second place gives him a decent three out of four chance. Loosing at RO16 means he is pretty unlikely to qualify with a mere 15% chance. How do his chances look like at RO8 and RO4 finishes?
|
very good!
and this is a good overview that the scores for different results within wcs are "broken". Look Innovation, even if he would have forfeited to play wcs s2 and s3, he still easily qualifiy for blizzcon. he had 4300 points before s2 begun.
|
I'd really be interested in knowing the effects of how other people do on qualification, especially for those players who are out of WCS.
I.e. if Scarlett goes out in ro32 wcs am, how does this effect Naniwa's chances of qualification (Scarlett is roughly 5 people still in WCS below nani, and it would help to know the overall chances of a foreigner at blizzcon).
|
Dingodile, what is wrong with that? Why shouldnt the winners of each season finale be eligible to compete at blizzcon?
|
|
|
|
|
|
EPT
