|
On October 07 2014 12:16 movac wrote:Show nested quote +On October 07 2014 05:17 Die4Ever wrote:On October 07 2014 05:06 Popkiller wrote:
yeah that's the problem I have with using aligulac for predictions.
why not just wait until someone is a lock (mathematically impossible for them to be eliminated from top 16), and until they are a lock, you can list the scenarios where they are eliminated. (I assume these lists would be huge, though) You can pretty much already do this with the events, you can see what secures them and what eliminates them, and you can see events that increase or decrease their chances. You can think of the chances as relative if you don't like aligulac ratings or use the all players are equal setting http://sc2.4ever.tv/?use_aligulac=0I do wish Aligulac weighted online matches less though, and tweaked the curve of rating changes when a strong player faces a weak player, the weak player farming is much too strong. Does a bo7 count more for aligulac than a best of 3? yeah aligulac would then have a lot less issue with foreign Koreans being rated so highly. as a result in twisted your own stats to say that jjakji to be 5th in chances to win blizzcon when in reality he's the weakest player out of the 16.
It's not that simple to just "weigh" something less. Trust me. We are continuing to work on it.
|
I think it is better without aligulac and other stuff next year. I dont like that the chance of x is higher than of y (both have same wcs points) to come to blizccon because of current performance. Just do "neutral" math. thats why we see Jaedong chance slight differently daily. 72% and next day 72.5% (numbers random chosen) while no sc2 happened/played there.
|
On October 07 2014 20:17 Grovbolle wrote:Show nested quote +On October 07 2014 12:16 movac wrote:On October 07 2014 05:17 Die4Ever wrote:On October 07 2014 05:06 Popkiller wrote:
yeah that's the problem I have with using aligulac for predictions.
why not just wait until someone is a lock (mathematically impossible for them to be eliminated from top 16), and until they are a lock, you can list the scenarios where they are eliminated. (I assume these lists would be huge, though) You can pretty much already do this with the events, you can see what secures them and what eliminates them, and you can see events that increase or decrease their chances. You can think of the chances as relative if you don't like aligulac ratings or use the all players are equal setting http://sc2.4ever.tv/?use_aligulac=0I do wish Aligulac weighted online matches less though, and tweaked the curve of rating changes when a strong player faces a weak player, the weak player farming is much too strong. Does a bo7 count more for aligulac than a best of 3? yeah aligulac would then have a lot less issue with foreign Koreans being rated so highly. as a result in twisted your own stats to say that jjakji to be 5th in chances to win blizzcon when in reality he's the weakest player out of the 16. It's not that simple to just "weigh" something less. Trust me. We are continuing to work on it.
To expand on this. We have tried some things to improve the rating algorithm and will continue to do so. However here are some points to reflect upon:
- Farming is not isolated to online cups.
- Farming is mostly a problem with newly added players. We have some ideas of how to solve this.
- Maybe your personal rating of players is wrong. People tend to overrate their favorites.
- Neither me or TheBB have a lot of time right now. But we are aware of this problem and it is brought up frequently in our internal discussions.
|
|
|
It would be fascinating to get some statistic for the skill with which aligulac predicts outcomes. Especially for specific players.
Also, I'm not trying to bash either the Blizzcon predictor or Aligulac; both are awesome and I look at them frequently. I just think they could be improved. It's good to hear that this is an issue that is being actively worked on.
|
On October 08 2014 03:59 e edgar wrote:
It would be fascinating to get some statistic for the skill with which aligulac predicts outcomes. Especially for specific players.
Also, I'm not trying to bash either the Blizzcon predictor or Aligulac; both are awesome and I look at them frequently. I just think they could be improved. It's good to hear that this is an issue that is being actively worked on.
Not sure what you're asking here. Do you want a percentage of correctly guessed scores?
|
Aligulac isn't perfect but I think its certainly more accurate than assuming a 50/50 result all the time. I do think some of the bias against their ratings is that sometimes aligulac doesnt value fan favorites as highly as people want because fans' conception of player skill tends to lag significantly behind current form. I do agree though that aligulac seems to overvalue 'foreign koreans. For example last I checked Sacsri was rated #3 despite not having achieved much of anything since his breakout tournament win a few months ago and with most of his recent recorded wins being against relatively less known foreigners.
|
On October 07 2014 22:45 Die4Ever wrote:Show nested quote +On October 07 2014 21:18 Dingodile wrote:
I think it is better without aligulac and other stuff next year. I dont like that the chance of x is higher than of y (both have same wcs points) to come to blizccon because of current performance. Just do "neutral" math. thats why we see Jaedong chance slight differently daily. 72% and next day 72.5% (numbers random chosen) while no sc2 happened/played there. Well you can view the neutral stats here http://sc2.4ever.tv/?use_aligulac=0[*]   Jaedong ( EG), is at ~ 62.5 %, Min WCS Points: 3200[*]  Pigbaby, is at ~ 12.5 %, Min WCS Points: 2500[*] HerO ( Liquid), is at ~ 12.5 %, Min WCS Points: 2250[*]  Scarlett ( Acer), is at ~ 12.5 %, Min WCS Points: 1950I really don't think it's that different since the matches left are between all strong players so they're pretty close. The Aligulac ratings mostly help with Dreamhacks where we all can guess the top 32 players with like 99% accuracy, it's kind of dumb to give a player like MMA a 50% chance to lose in the round of 64 there. Next year I might try adjusting it so near equal players have closer to 50/50 but keep the discrepancy when the players are far apart in Aligulac ratings. And this link will be easier to find next year.
Please keep the Aligulac prediction column and just add another "unbiased" one if you want. I think the way it's done at the moment is exactly how it should be! Assuming 50% winrate, even between top players, is a mistake. I'd rather rely on the science behind Aligulac, which makes a lot of sense.
|
Aligulac is not science... it is probabilities based on performances in the past.
|
On October 08 2014 21:43 gneGne wrote:
Aligulac is not science... it is probabilities based on performances in the past.
So you mean to tell me that mathematics (statistics and probabilities are parts of maths) are not a science? The way Aligulac proceeds to compute their predictions is as close to science as you'll get from sc2 data. (read their FAQ)
In any case, I'd like to hear the arguments defending the point that 50/50 would actually be a better prediction for top games... This is just wrong. Feelings are biased. "The good thing about science is that it's true whether you believe it or not."
|
nowadays everything is called a science ^^
|
On October 08 2014 23:16 CursedFeanor wrote:Show nested quote +On October 08 2014 21:43 gneGne wrote:
Aligulac is not science... it is probabilities based on performances in the past. So you mean to tell me that mathematics (statistics and probabilities are parts of maths) are not a science? The way Aligulac proceeds to compute their predictions is as close to science as you'll get from sc2 data. (read their FAQ) In any case, I'd like to hear the arguments defending the point that 50/50 would actually be a better prediction for top games... This is just wrong. Feelings are biased. "The good thing about science is that it's true whether you believe it or not."
Statistics are great and all, but aligulac has a problem where it reward players who play against weaker opponents and win a lot over players who play against and win less against stronger opponents. If two players were completely even, but one played in Korea and the other NA, the NA player would likely be rated higher because they win more, but that doesn't mean it's an accurate rating.
|
You guys are all discussing Aligulac ratings in terms of great players like Yoda vs Innovation, but you're forgetting that where they make the most difference is when it's RandomUSAPlayer vs Polt in WCS AM qualifiers, or RandomOpenBracketPlayer vs MMA at a Dreamhack, these matches are a big deal for Blizzcon chances and doing them 50/50 throws it all off especially in the beginning of the year. In the end of the year now we only have some top players facing each other (and there is an option to view the 50/50 stats if you want), but in the beginning of the year there'll be a lot of those weaker players vs top players matches that need to be predicted when simulating 5 Dreamhacks and a few IEMs with open brackets, and WCS qualifiers, and every other tournament of the year.
|
On October 08 2014 23:23 sharkie wrote:
nowadays everything is called a science ^^
that's because science is a general term which can encompass many different disciplines? and it's not "nowadays" it's always been true. if you use the scientific method you're doing science
|
On October 08 2014 21:03 CursedFeanor wrote:Show nested quote +On October 07 2014 22:45 Die4Ever wrote:On October 07 2014 21:18 Dingodile wrote:
I think it is better without aligulac and other stuff next year. I dont like that the chance of x is higher than of y (both have same wcs points) to come to blizccon because of current performance. Just do "neutral" math. thats why we see Jaedong chance slight differently daily. 72% and next day 72.5% (numbers random chosen) while no sc2 happened/played there. Well you can view the neutral stats here http://sc2.4ever.tv/?use_aligulac=0[*] Jaedong ( EG), is at ~ 62.5 %, Min WCS Points: 3200[*] Pigbaby, is at ~ 12.5 %, Min WCS Points: 2500[*] HerO ( Liquid), is at ~ 12.5 %, Min WCS Points: 2250[*] Scarlett ( Acer), is at ~ 12.5 %, Min WCS Points: 1950I really don't think it's that different since the matches left are between all strong players so they're pretty close. The Aligulac ratings mostly help with Dreamhacks where we all can guess the top 32 players with like 99% accuracy, it's kind of dumb to give a player like MMA a 50% chance to lose in the round of 64 there. Next year I might try adjusting it so near equal players have closer to 50/50 but keep the discrepancy when the players are far apart in Aligulac ratings. And this link will be easier to find next year. Please keep the Aligulac prediction column and just add another "unbiased" one if you want. I think the way it's done at the moment is exactly how it should be! Assuming 50% winrate, even between top players, is a mistake. I'd rather rely on the science behind Aligulac, which makes a lot of sense.
I think it would be too confusing to show some 50/50 stats along with Aligulac based stats at the same time, reading the Median WCS Points, and Upcoming Match previews, and events. I'm just going to maybe make it slightly easier to find a link to the 50/50 stats so you can view those if you really want, but they're pretty useless until the end of the year.
|
On October 08 2014 23:16 CursedFeanor wrote:Show nested quote +On October 08 2014 21:43 gneGne wrote:
Aligulac is not science... it is probabilities based on performances in the past. So you mean to tell me that mathematics (statistics and probabilities are parts of maths) are not a science? The way Aligulac proceeds to compute their predictions is as close to science as you'll get from sc2 data. (read their FAQ) In any case, I'd like to hear the arguments defending the point that 50/50 would actually be a better prediction for top games... This is just wrong. Feelings are biased. "The good thing about science is that it's true whether you believe it or not."
I dont understand some of the logic on Aligulac. For example, Sacsri won a unimpressive dreamhack , beating most foreigners and EU koreans and his rating jumped from 1500 to 2100....meanwhile, a 4 times GSL finalist soo barely hit 2000 mark and during his GSL runs he was ranked anywhere from 1700 to 1800. This does not make any sense .. at all....
|
On October 09 2014 01:49 cpower wrote:Show nested quote +On October 08 2014 23:16 CursedFeanor wrote:On October 08 2014 21:43 gneGne wrote:
Aligulac is not science... it is probabilities based on performances in the past. So you mean to tell me that mathematics (statistics and probabilities are parts of maths) are not a science? The way Aligulac proceeds to compute their predictions is as close to science as you'll get from sc2 data. (read their FAQ) In any case, I'd like to hear the arguments defending the point that 50/50 would actually be a better prediction for top games... This is just wrong. Feelings are biased. "The good thing about science is that it's true whether you believe it or not." I dont understand some of the logic on Aligulac. For example, Sacsri won a unimpressive dreamhack , beating most foreigners and EU koreans and his rating jumped from 1500 to 2100....meanwhile, a 4 times GSL finalist soo barely hit 2000 mark and during his GSL runs he was ranked anywhere from 1700 to 1800. This does not make any sense .. at all....
It must favor champions over finalists by a very large margin. Other than that, the only thing to keep in mind is that soO's first GSL finals was last year, so as far as 2014 WCS is concerned, he was only a two-time runner-up until he faced Innovation. Still, two time runner-up of GSL should mean a hell of a lot more than it does. There's no way that soO should not have more WCS points than anybody else in the system in 2014. There just isn't.
|
On October 09 2014 02:32 pure.Wasted wrote:Show nested quote +On October 09 2014 01:49 cpower wrote:On October 08 2014 23:16 CursedFeanor wrote:On October 08 2014 21:43 gneGne wrote:
Aligulac is not science... it is probabilities based on performances in the past. So you mean to tell me that mathematics (statistics and probabilities are parts of maths) are not a science? The way Aligulac proceeds to compute their predictions is as close to science as you'll get from sc2 data. (read their FAQ) In any case, I'd like to hear the arguments defending the point that 50/50 would actually be a better prediction for top games... This is just wrong. Feelings are biased. "The good thing about science is that it's true whether you believe it or not." I dont understand some of the logic on Aligulac. For example, Sacsri won a unimpressive dreamhack , beating most foreigners and EU koreans and his rating jumped from 1500 to 2100....meanwhile, a 4 times GSL finalist soo barely hit 2000 mark and during his GSL runs he was ranked anywhere from 1700 to 1800. This does not make any sense .. at all.... It must favor champions over finalists by a very large margin. Other than that, the only thing to keep in mind is that soO's first GSL finals was last year, so as far as 2014 WCS is concerned, he was only a two-time runner-up until he faced Innovation. Still, two time runner-up of GSL should mean a hell of a lot more than it does. There's no way that soO should not have more WCS points than anybody else in the system in 2014. There just isn't.
Aligulac doesn't know whether or not a match was for the championship, just who the players are and their ratings.
As for the quoted post and soO's score, you can click on the arrows to the right for each time period in Soo's historical rating to see who he played, what each of their ratings were beforehand, and how soO's rating was affected.
The short answer to the soO question is that soO tended to lose a lot of non-GSL matches during time periods where he won a ton of GSL matches (e.g., his first finals run was interspersed with non-GSL losses to Byul, Maru, San, and Bunny, not to mention two GSL group stage losses). Sacsri only lost one match during his huge jump period, and the players he beat had generally similar ratings to the ones that soO tends to beat in proleague.
|
On October 09 2014 07:37 frogrubdown wrote:Show nested quote +On October 09 2014 02:32 pure.Wasted wrote:On October 09 2014 01:49 cpower wrote:On October 08 2014 23:16 CursedFeanor wrote:On October 08 2014 21:43 gneGne wrote:
Aligulac is not science... it is probabilities based on performances in the past. So you mean to tell me that mathematics (statistics and probabilities are parts of maths) are not a science? The way Aligulac proceeds to compute their predictions is as close to science as you'll get from sc2 data. (read their FAQ) In any case, I'd like to hear the arguments defending the point that 50/50 would actually be a better prediction for top games... This is just wrong. Feelings are biased. "The good thing about science is that it's true whether you believe it or not." I dont understand some of the logic on Aligulac. For example, Sacsri won a unimpressive dreamhack , beating most foreigners and EU koreans and his rating jumped from 1500 to 2100....meanwhile, a 4 times GSL finalist soo barely hit 2000 mark and during his GSL runs he was ranked anywhere from 1700 to 1800. This does not make any sense .. at all.... It must favor champions over finalists by a very large margin. Other than that, the only thing to keep in mind is that soO's first GSL finals was last year, so as far as 2014 WCS is concerned, he was only a two-time runner-up until he faced Innovation. Still, two time runner-up of GSL should mean a hell of a lot more than it does. There's no way that soO should not have more WCS points than anybody else in the system in 2014. There just isn't. Aligulac doesn't know whether or not a match was for the championship, just who the players are and their ratings. As for the quoted post and soO's score, you can click on the arrows to the right for each time period in Soo's historical rating to see who he played, what each of their ratings were beforehand, and how soO's rating was affected. The short answer to the soO question is that soO tended to lose a lot of non-GSL matches during time periods where he won a ton of GSL matches (e.g., his first finals run was interspersed with non-GSL losses to Byul, Maru, San, and Bunny, not to mention two GSL group stage losses). Sacsri only lost one match during his huge jump period, and the players he beat had generally similar ratings to the ones that soO tends to beat in proleague.
The last ten opponents beat soO were: Innovation Solar Bbyong Super Classic Flash Hydra herO True Rogue
The last ten opponents beat Sacsri were: ShoWTime Hyun Snute First herO Elazer Hurricane RagnaroK Lilbow Zoun
I dont think I need to say more, the list explains itself.
|
On October 12 2014 00:50 cpower wrote:Show nested quote +On October 09 2014 07:37 frogrubdown wrote:On October 09 2014 02:32 pure.Wasted wrote:On October 09 2014 01:49 cpower wrote:On October 08 2014 23:16 CursedFeanor wrote:On October 08 2014 21:43 gneGne wrote:
Aligulac is not science... it is probabilities based on performances in the past. So you mean to tell me that mathematics (statistics and probabilities are parts of maths) are not a science? The way Aligulac proceeds to compute their predictions is as close to science as you'll get from sc2 data. (read their FAQ) In any case, I'd like to hear the arguments defending the point that 50/50 would actually be a better prediction for top games... This is just wrong. Feelings are biased. "The good thing about science is that it's true whether you believe it or not." I dont understand some of the logic on Aligulac. For example, Sacsri won a unimpressive dreamhack , beating most foreigners and EU koreans and his rating jumped from 1500 to 2100....meanwhile, a 4 times GSL finalist soo barely hit 2000 mark and during his GSL runs he was ranked anywhere from 1700 to 1800. This does not make any sense .. at all.... It must favor champions over finalists by a very large margin. Other than that, the only thing to keep in mind is that soO's first GSL finals was last year, so as far as 2014 WCS is concerned, he was only a two-time runner-up until he faced Innovation. Still, two time runner-up of GSL should mean a hell of a lot more than it does. There's no way that soO should not have more WCS points than anybody else in the system in 2014. There just isn't. Aligulac doesn't know whether or not a match was for the championship, just who the players are and their ratings. As for the quoted post and soO's score, you can click on the arrows to the right for each time period in Soo's historical rating to see who he played, what each of their ratings were beforehand, and how soO's rating was affected. The short answer to the soO question is that soO tended to lose a lot of non-GSL matches during time periods where he won a ton of GSL matches (e.g., his first finals run was interspersed with non-GSL losses to Byul, Maru, San, and Bunny, not to mention two GSL group stage losses). Sacsri only lost one match during his huge jump period, and the players he beat had generally similar ratings to the ones that soO tends to beat in proleague. The last ten opponents beat soO were: Innovation Solar Bbyong Super Classic Flash Hydra herO True Rogue The last ten opponents beat Sacsri were: ShoWTime Hyun Snute First herO Elazer Hurricane RagnaroK Lilbow Zoun I dont think I need to say more, the list explains itself.
I took you to be making a different point than the old one that aligulac seems to overrate players who play mostly outside of Korea. Instead I took you to be asking how even by that standard soO would never experience as big a rating jump as sacsri did given all soO's success, and I answered that question. Apparently you were making the first point.
|
|
|
|
|
|
EPT
