EPT[Aligulac June 12 list] Your ratings are Bogus!!!! - Page 3
| Forum Index > SC2 General |
Shellshock
United States97276 Posts
| ||
|
gobbledydook
Australia2603 Posts
On June 13 2013 13:27 BisuDagger wrote: Yes but he made a huge mistake in his blog by forgetting point value. As in I unlike the machine can take risks on Soulkey and gain 7 points while the machine has to now get 7 1 point votes correct to catch up. So while I'm at 60% I'm still doing really well with the high value votes That is so wrong 'take risks on Soulkey and gain 7 points...' that completely doesn't make sense, point value has no relation to how likely you are to be correct because whether it's worth 7 points or 1 it's still a 50-50 bet. | ||
|
Rescawen
Finland1028 Posts
| ||
Zealously
East Gorteau22261 Posts
| ||
|
Grovbolle
Denmark3805 Posts
On June 13 2013 08:08 Emzeeshady wrote: Lol, the GSL champion and WCS world finals semi finalist isn't in the top ten... He is only 10 points from top 10 though On June 13 2013 08:09 Serimek wrote: Who's flashpoint at 84th rank ?_? A statistical anomaly.  On June 13 2013 11:16 leova wrote: wow....for real? that's, um....kinda silly, to be honest, and borderline racist Why is someone automatically a better Video Game Player simply based on where they were born? Being born in Korea makes you 1% cyborg, so you're auto-better than a guy from Brazil or France or Ethiopia? good lord, no wonder these ratings are stupid Who do you fear more?   |||||||||||| ||||||||||||The Koreans in our DB were on average more than 200 points higher rated than foreigners, so we sped up the point transfer a bit On June 13 2013 13:05 wklbishop wrote: And this isn't really much of a feedback since I haven't looked into the actual code/math of it. But has theBB ever ran his prediction system against historical matches or even liquibet system? Maybe you have a 1 and subtract it by whoever won's pre-game chances. Rinse and repeat 1000 or so times. Then you have a relative benchmark that you can run as a test for future improvements as actually being improvements. Sort of like RMSE and trying to decrease it as much as possible, but in this case we don't really need to square and sqrt. That way we might be able to have a way of measuring how good the system is currently and a way to check the continuing progress and instead of giving 200 points randomly, we can give points that would go to optimizing the RMSE. Once again, I haven't actually looked in depth into this, so feel free to ignore my suggestion. Using the data in the DB to compare actual results with predicted results. After period 1, the model try to predict all the results entered in period 2. After period 2, the model try to predict all the results entered in period 3, using the results from period 1 & 2. Rinse and repeat. The models parameters are being optimized trying to raise the predictabillity. On June 13 2013 16:47 Zealously wrote: Dreamhack/MLG/OSL will give Life back the #1 in no time (or actually 4 weeks, but yeah) Most likely after this weekend. On June 13 2013 Grovbolle wrote: Now before we get into this, remember that a difference of around 250 or less means that there is no statistically significant difference between two players, but who am I kidding, you just want to rage and yell at our stats, so here we go. Most people seems to miss this in the OP | ||
|
Serimek
France2274 Posts
That, I understood  But what I meant is who is he in real life ? Just a random master player who get to later stages of an online cup by sheer luck ? Or maybe an up-and-coming player no one noticed before ? | ||
|
Koshi
Belgium38799 Posts
| ||
|
TheBB
Switzerland5133 Posts
On June 13 2013 11:16 leova wrote: that's, um....kinda silly, to be honest, and borderline racist Why is someone automatically a better Video Game Player simply based on where they were born? Being born in Korea makes you 1% cyborg, so you're auto-better than a guy from Brazil or France or Ethiopia? I was loathe to try it for a good while, but I've warmed up to the idea. In the old system, it had taken the Koreans 3 years to gain a mean rating advantage on the foreign scene of about 140 points. When Koreans went abroad and participated in foreign tournaments, they consistently overperformed by several hundred points. When ForGG went to France he gained almost 500 points in only 3-4 months. Assuming his real skill didn't change too much over that time, that means he was underrated compared to the Europeans by at least 300 points. If I had left the system untouched, it would have taken something like 5 or 6 years until the ratings of Koreans had caught up to their actual level. That is unacceptable. It's a problem that you don't have in chess and other places where such ratings exist. I didn't have any papers to draw on, so I just came up with the best idea I could. It's definitely not racism. I don't subscribe to the idea that a higher Aligulac rating makes someone a more valuable human being. As for the offline weighing, I was also not willing to try it for a long time, but it also helps the Korean scene. You see, Koreans consistently play fewer games than foreigners do. This means that while the foreign ranking shows an accurate picture of today, the Korean ranking shows an accurate picture of two months back. That's also not acceptable. However, while Koreans do not play very often, when they do, they almost always play offline. So I consider the offline boost a way to speed up the inter-Korean adjustments, and the Korean boost a way to speed up the Korean-foreigner adjustments. There are two ways you can view ratings. You can consider it as a way to quantify actual real skill, or you can consider it a competition unto itself. I like both, but in Starcraft 2 I don't think it's possible to accommodate both viewpoints. If I make it fair like a competition, it will be a poor indicator of skill (it will claim Koreans are poorer than they actually are). With this change, I have decided to move more in the former direction. It's a better indicator of actual skill, but it cannot any longer be considered a "competition" since the rules are no longer fair. Well, so be it. | ||
|
Tsubbi
Germany7979 Posts
| ||
|
TheBB
Switzerland5133 Posts
On June 13 2013 19:28 Tsubbi wrote: lol how much will innovation lose after shine 6 pooled him ? Ha, just for you I made an extraordinary rating update. The answer is 15 points.  http://aligulac.com/players/48-INnoVation/period/87/ Edit: The shoutout to MasterOfPuppets and Shellshock is well deserved! | ||
|
Tsubbi
Germany7979 Posts
| ||
|
Sabre
United Kingdom1086 Posts
| ||
|
docvoc
United States5491 Posts
| ||
|
LimitSEA
Australia9580 Posts
| ||
|
FCReverie
Australia103 Posts
EDIT: It has actually gotten worse. Now it is struggling to load the page at all. | ||
BisuDagger
Bisutopia19231 Posts
On June 13 2013 16:34 gobbledydook wrote: That is so wrong 'take risks on Soulkey and gain 7 points...' that completely doesn't make sense, point value has no relation to how likely you are to be correct because whether it's worth 7 points or 1 it's still a 50-50 bet. What are you talking about? First off if one person has an 80% chance to win a game then even if it is between two people it doesn't mean it's 50/50, it's 80/20. People are not coins. It's not black and white/heads or tails otherwise the average person would have a 50% LB ratio. And a lot of the time it's predicting two people to make it through a group so that's an added variable to take into consideration. If you truly believe that MVP vs Idra is a 50/50 vote then don't reply because we will never see eye to eye on this. But your other statement that points don't make a difference in how right I am is true. But that wasn't my point so pay attention. The blog was about using the computer to make all your LB choices for you and that you can stay at the top using Aligulac. So if Innovation was favored in the computer then you lost 7 points by trusting the computer. I took a risk and went with Soulkey and it paid off. So my point is I can actually having a lower correct voting % then Aligulac but still have more points because I can take into all factors when choosing high point value votes. example: If Aligulac gets 7/10 votes correct and I got 6/10 correct. Does this mean I'm losing? No, not if my point total is higher then Aligulac. | ||
|
slowbacontron
United States7722 Posts
On June 14 2013 01:12 FCReverie wrote: More than 1/2 the time when I try to use the search function the page doesn't load (I get an Oops Chrome couldn't connect message) wonder if that is just me. EDIT: It has actually gotten worse. Now it is struggling to load the page at all. Looks like it's down right now, I'd wait to see if it gets better. | ||
|
TheBB
Switzerland5133 Posts
On June 14 2013 01:29 slowbacontron wrote: Looks like it's down right now, I'd wait to see if it gets better. I don't know shit about running a web server. I just made a site... not my fault people are using it. xD I ordered a reboot, let's see what happens. Update 19:13 CEST: No news so far, the VPS dashboard is bugged too. I've submitted a ticket to the host, so that's all for the moment. Update 19:37 CEST: The host's websites are also not responding, now. Update 20:26 CEST: Service appears back to normal. | ||
|
mnck
Denmark1518 Posts
| ||
|
scypio
Poland2127 Posts
On June 14 2013 03:55 mnck wrote: Life will return to number 1 np Yeah, he's going to Europe for a point farming fest. | ||
| ||
