EPTBlizzcon qualification probabilities simulation - Page 4
| Forum Index > SC2 General |
|
Gojii
Belgium317 Posts
| ||
|
AySz88
United States83 Posts
On September 13 2013 21:12 Die4Ever wrote: I may switch to using relative %s, if you guys can think of a good way to word the text for the events that would be helpful  I would suggest "T-9th" or "T9-16th" instead of "16th place", because we don't know whether 16th means "made it to round of 16" (aka top 16) or "eliminated in round of 16" (aka T-9th). (Or maybe I'm confused and you did actually mean top 16?) | ||
|
Die4Ever
United States17738 Posts
On September 14 2013 04:14 AySz88 wrote: I would suggest "T-9th" or "T9-16th" instead of "16th place", because we don't know whether 16th means "made it to round of 16" (aka top 16) or "eliminated in round of 16" (aka T-9th). (Or maybe I'm confused and you did actually mean top 16?) It means they were eliminated at that place. So if it says 16th place it means they lost in the round of 16 in that instance. | ||
|
SolidMoose
United States1240 Posts
But seriously, this is really cool. Nice to see the chances from the TL article have some numbers to them now | ||
|
Steel
Japan2283 Posts
Quick questions: -Did you use ELO or winrates? -Did you use matchup specific ELO (or winrates) or just general ELO (or winrates)? -If you used ELO, did you use the uncertainties quoted by Aligulac? -If so, what does the uncertainty look like for 300000 runs? I'm just curious, it should be really small. Besides the Aligulac uncertainties are very fishy. Example: ELO 1871 ± 81 for Taeja at 1182 games vs 1833 ± 89 for sOs at 247 games. Low number of games + consistent results ~ High number of games + inconsistent results. I get it, but I don't like it. Anyway, good work, really interesting stuff. | ||
|
Boucot
France15997 Posts
| ||
Cracy
Poland221 Posts
On September 14 2013 04:08 Anomi wrote: It would actually take me allot of time or maybe impossible to figure out on how to create a model that would fill this roll. . Using tools as http://www.r-project.org/ that is free will give you to tools to maybe create a better simulation . Using a logistical model to work with this kind of data might work better but not sure about that (http://en.wikipedia.org/wiki/Logistic_regression). For now you should just add a similar response from lolfail9001 and state what those cases are “Aligulac just provides CHANCES for player to advance from group that he uses in his simulation to calculate chances of player advancing or not in certain cases.” It’s still not 100% correct way of saying it but people would probably just be confused if you tried to explain that the chance from Aligulac ranking is a estimation of what the chance can be and not the actual chance the player that is then used in a simulation to predict what the actual chance could be to qualify to Blizcon . The bottom line is that there is no wrong or right as long as you state the assumption the model is based on and the reasoning behind the decision you made(for instance why using Aligulac ranking as factor and not win ratio ect ). This is the only thing I would actually tell you that needs to be done better. PS:Sry if i sounded critical is still a nice contribution and well done work Why would you even write all the posts? I do understand the constructive criticism but hey... You are now offending the intelligence of all the people who read it and ENJOY IT. For one thing I work with stats and models for living and still it wouldn't come to my mind to criticize the work done. | ||
|
ogsgodlike
Bulgaria31 Posts
| ||
|
Die4Ever
United States17738 Posts
On September 14 2013 04:26 Steel wrote: Really cool, and while monte carlo simulations are solid, I very much doubt the precision of Aligulac ratings. Quick questions: -Did you use ELO or winrates? -Did you use matchup specific ELO (or winrates) or just general ELO (or winrates)? -If you used ELO, did you use the uncertainties quoted by Aligulac? -If so, what does the uncertainty look like for 300000 runs? I'm just curious, it should be really small. Besides the Aligulac uncertainties are very fishy. Example: ELO 1871 ± 81 for Taeja at 1182 games vs 1833 ± 89 for sOs at 247 games. Low number of games + consistent results ~ High number of games + inconsistent results. I get it, but I don't like it. Anyway, good work, really interesting stuff. I use the aligulac rating which is similar to Elo. Currently I'm just using the general aligulac ratings and not doing it matchup specific, but I definitely plan to soon I am not currently using the uncertainties Thanks! | ||
|
Noonius
Estonia17413 Posts
as someone who likes numbers and statistics, this is like heaven | ||
|
KoRStarvid
Sweden767 Posts
On September 14 2013 04:26 Steel wrote: Really cool, and while monte carlo simulations are solid, I very much doubt the precision of Aligulac ratings. Quick questions: -Did you use ELO or winrates? -Did you use matchup specific ELO (or winrates) or just general ELO (or winrates)? -If you used ELO, did you use the uncertainties quoted by Aligulac? -If so, what does the uncertainty look like for 300000 runs? I'm just curious, it should be really small. Besides the Aligulac uncertainties are very fishy. Example: ELO 1871 ± 81 for Taeja at 1182 games vs 1833 ± 89 for sOs at 247 games. Low number of games + consistent results ~ High number of games + inconsistent results. I get it, but I don't like it. Anyway, good work, really interesting stuff. I don't understand how ELO is related to this at all. Is Glicko based on ELO, or am I missing something else here? Aligulac rankings seem pretty good imo. Are there any better alternatives? I really enjoyed scrolling through the results of this simulation! And unlike some of the other posters, I think the model is appropriately complex! :-) | ||
|
Mezox
Sweden6 Posts
I am a math/statistics student myself and I'm a big fan of stuff like this! | ||
|
TaishiCi
Korea (South)211 Posts
Only Scarlett and Naniwa have actual chance to spoil the perfect goodbye to primetime sc2. | ||
|
lolfail9001
Russian Federation40190 Posts
On September 14 2013 05:21 TaishiCi wrote: Would love to have it be all Korean Finals for the Annual Finals. Only Scarlett and Naniwa have actual chance to spoil the perfect goodbye to primetime sc2. Hm. Nice bait. Either way, nice work OP, i wonder how many time it took to simulate it all. | ||
|
Anomi
Sweden149 Posts
On September 14 2013 04:27 Cracy wrote: Why would you even write all the posts? I do understand the constructive criticism but hey... You are now offending the intelligence of all the people who read it and ENJOY IT. For one thing I work with stats and models for living and still it wouldn't come to my mind to criticize the work done. I did say sry realising the posts i wrote was a bit to critical and to some extent unnecessary. As for why? the same reason you posting a critic on someone opinion .Just to clarify the work is good and I am just over critical. | ||
|
BaneRiders
Sweden3630 Posts
 | ||
|
xN.07)MaK
Spain1159 Posts
Great work! | ||
|
ffadicted
United States3545 Posts
This post is absolutely awesome, thanks for the work tbh, I was actually trying to figure out myself if my boy Rain could make it haha | ||
|
Die4Ever
United States17738 Posts
On September 14 2013 05:22 lolfail9001 wrote: Hm. Nice bait. Either way, nice work OP, i wonder how many time it took to simulate it all. Thanks. It takes about 40 minutes to run it with 300,000 samples like in the OP. I could optimize it to make it faster, but I haven't spent the time to do it and don't really need it to be faster anyways. | ||
|
Die4Ever
United States17738 Posts
On September 14 2013 05:29 BaneRiders wrote: This is very cool! I would like to see the most probable top 16 though, the players that will participate in the end. This list could be updated (as you like) and have the, uh, I don't know, the 4 closest to take a spot as well or something. Would be cool to see how this top 20 will develop. Thanks. I'm not sure I understand what your suggestion is though. Could you maybe elaborate? Also Blizzcon is 16 players not 20. | ||
| ||
