EPTnice analysis but completly worthless
User was temp banned for this post.
SC2 Ladder Analysis: Division Tiers - Page 5
| Forum Index > SC2 General |
|
TehForce
1072 Posts
nice analysis but completly worthless User was temp banned for this post. | ||
|
Shadowed
United States679 Posts
| ||
|
aristarchus
United States652 Posts
 I don't think the division differences should be that surprising. Think about it from Blizzard's perspective. Why did they want the division/league system at all, instead of one big ladder? Clearly they decided that the division system would motivate people to play more. With 100 people in your division, you can see the people you're competing against and fight to move a couple spots. If you're #10487 and then after winning you're #10224 it's hard to be that motivated from it (or at least, that was the bet Blizzard clearly made). If it's just trivial to convert to your actual ladder ranking, it somewhat defeats the purpose. They want you to care about your division ranking, not just your overall ranking. Obviously the logic doesn't really apply to the very top players, which is probably part of why they're creating a new higher league, where I bet the overall ranking will be quite clear and easy to see. We don't know if these division bonus values change over time. It's possible that they just throw people into divisions randomly, and then add the bonus values to try to create the impression that diamond divisions don't differ "too much" in skill. It might be that if a division's players got on average a lot better, the bonus value would be adjusted. (I'm not sure how a change would actually be implemented, since they can't just have you randomly jump a bunch of points without you noticing, but I could imagine more subtle ways of doing it.) | ||
|
smurfzg
Sweden8 Posts
On November 18 2010 11:38 starcraft2leverage wrote: Everyone gains and loses points the same from wins and losses, the difference is the starting point when you join diamond. You got an extra 300 or so points when you joined, others got 63 or none or whatevs. Remember that you lose a seemingly arbitrary amount of points when you get promoted to a new league. See http://www.teamliquid.net/forum/viewmessage.php?topic_id=163057 It seems likely that your points get adjusted depending on what class your platinum division was compared to your diamond division (in addition to some penalty for it being diamond instead of platinum). It could be worth investigating. | ||
Excalibur_Z
United States12235 Posts
On November 19 2010 03:11 smurfzg wrote: Remember that you lose a seemingly arbitrary amount of points when you get promoted to a new league. See http://www.teamliquid.net/forum/viewmessage.php?topic_id=163057 It seems likely that your points get adjusted depending on what class your platinum division was compared to your diamond division (in addition to some penalty for it being diamond instead of platinum). It could be worth investigating. That's one thing I was considering, too. It's too bad that thread is so full of estimations instead of reliable numbers, or we might be able to gain some important information from it. I know that when I was promoted, my point change was -63. Could be a coincidence, could be that I moved from the highest Plat division tier to the lowest Diamond (suggesting one contiguous ladder), could be that I moved from a Plat division to a Diamond division one tier lower (suggesting completely separate league point values). If anyone has solid numbers to contribute (I think Mendelfist does) I'd love to see them. | ||
|
Mendelfist
Sweden356 Posts
On November 19 2010 03:22 Excalibur_Z wrote: That's one thing I was considering, too. It's too bad that thread is so full of estimations instead of reliable numbers, or we might be able to gain some important information from it. I know that when I was promoted, my point change was -63. Could be a coincidence, could be that I moved from the highest Plat division tier to the lowest Diamond (suggesting one contiguous ladder), could be that I moved from a Plat division to a Diamond division one tier lower (suggesting completely separate league point values). If anyone has solid numbers to contribute (I think Mendelfist does) I'd love to see them. I still update my match history sometimes if thats what you are looking for: http://www.lysator.liu.se/~john/history.html It only contains one demotion and one promotion though. | ||
|
Excalibur_Z
United States12235 Posts
On November 19 2010 04:02 Mendelfist wrote: I still update my match history sometimes if thats what you are looking for: http://www.lysator.liu.se/~john/history.html It only contains one demotion and one promotion though. So the promotion took you from 1164 down to 959 which is a difference of 205. You gained +32 points in your promotion match. 205 is 16 over 189 which is a multiple of 63. Maybe it ate your 16 from bonus pool like it did for me, but didn't apply it? Or it could be unrelated. I do see that your subsequent game's point value is bugged, however. When I was promoted, I gained 20+20 points for my last win, but 40 points were deducted from my bonus pool instead of 20. I lost 63 points when I got promoted. Again, maybe a coincidence. Your demotion was from 48 to 144, a difference of 96. You lost 18 points in your demotion match. I don't see any correlation here at all, unfortunately. | ||
|
chadissilent
Canada1187 Posts
This is why divisions are meaningless. Ice cream sales are up in the summer. Reported cases of rape are also up in the summer. Does this mean that eating ice cream causes you to be raped? | ||
|
Shadowed
United States679 Posts
The exact mechanics of how divisions are spun up can't really be proven either way right now. | ||
|
KaienFEMC
Canada127 Posts
In Taiwan server, the bottom top 200 players usually are people with very few games due to the small player pool. There are even platinum players who made it to the top 200 list (No. 198 30 wins with 4 loses). How would you explain this from your analysis/algorithm? | ||
Heyoka
Katowice25012 Posts
Would it be helpful to track the same data for eu/kr as well to try to find that answer? Do you have a rough idea of how many diamond divisions exist in total right now? | ||
|
Shadowed
United States679 Posts
On November 19 2010 05:22 heyoka wrote: This is awesome, great work. How close are we to being able to compare across divisions as a whole? ie getting an absolute rating for people not in the top 200. Would it be helpful to track the same data for eu/kr as well to try to find that answer? Do you have a rough idea of how many diamond divisions exist in total right now? Never :p The top 200 covers rankings for 4,000 to 6,000 people per region depending on division variety. Will be able to do cross division rankings for the Masters and Grand Masters leagues, but those won't have divisions. | ||
|
Excalibur_Z
United States12235 Posts
On November 19 2010 05:15 KaienFEMC wrote: I only have one question. In Taiwan server, the bottom top 200 players usually are people with very few games due to the small player pool. There are even platinum players who made it to the top 200 list (No. 198 30 wins with 4 loses). How would you explain this from your analysis/algorithm? The Top 200s in the past (weeks 1 and 2 especially) even on the NA server have included some Platinum players. If we had that data, that could potentially be very useful in determining how points can be compared across leagues as well. I'll look into it. | ||
|
darmousseh
United States3437 Posts
What the OP more likely found was a product of the algorithm that they use to sort, most likely a product of recent activity. It has been observed that more inactive players tend to get lower on the ranks as opposed to active players. This might be a coincidence also since 63 is divisble by 7, but otherwise there is no reason for 63 to be this magic #. The best way to verify this is to wait til next season and compare results. Final Note: Blizzard, freaking just make the system simple. Show us our moving MMR. You can even subtract 2* sigma from our mean to make it easy for us to understand. We want to know our rating, not some point thing. | ||
|
Shadowed
United States679 Posts
On November 19 2010 05:33 Excalibur_Z wrote: The Top 200s in the past (weeks 1 and 2 especially) even on the NA server have included some Platinum players. If we had that data, that could potentially be very useful in determining how points can be compared across leagues as well. I'll look into it. You know now that I think about it, going to be screwed come next season with Grand Masters and Masters. Won't have anything to really reference to figure out what division is what unless the MMR formula is derived from points, which is essentially impossible. | ||
|
YokaY
United States108 Posts
It could explain why some extremely high tier players would stay in plat for an extremely long time. If the divisions were separated by skill within the league, they may have to wait a lot longer for enough high skill players to fill create a new division. | ||
|
Eeryck
United States184 Posts
On November 19 2010 05:34 darmousseh wrote: There is a fundamental flaw with this understanding. When assigning people divisions, how would it know to assign to one division or another? What makes 1 player being promoted from plat to diamond better than a different player? How would it know to create a new "S" class division. What i think is more likely is that good players got promoted in the system a lot faster than lower ranked players, therefore there are a few divisions that have significantly more pros than other divisions. If a player on some random diamond division was to suddenly become better, he would never switch divisions and therefore you could say "his division must be good, look they have someone in the top 5". What the OP more likely found was a product of the algorithm that they use to sort, most likely a product of recent activity. It has been observed that more inactive players tend to get lower on the ranks as opposed to active players. This might be a coincidence also since 63 is divisble by 7, but otherwise there is no reason for 63 to be this magic #. The best way to verify this is to wait til next season and compare results. Final Note: Blizzard, freaking just make the system simple. Show us our moving MMR. You can even subtract 2* sigma from our mean to make it easy for us to understand. We want to know our rating, not some point thing. It won't create new S-Class divisions, but for someone that is new to diamond to crack the top 200 they would need to have significantly higher point total because of the division modifier. What this analysis says is that all 2000 pt diamond players are not equal or necessarily even close in MMR. If someone is really progressing quickly, it may also use a demotion to free up a spot in one of the higher level S-Class divisions to move that person into the appropriate level diamond division. This has good correlation with some high level players taking an exceedingly long time to get to diamond. They had to wait for a S-Class league demotion to make a spot for them. Points are only utilized for relative rank in your own division. If you want to know rank compared to other players in different divisions you need this division modifier. This topic is hard to accept because there was a general consensus that x-level diamond was = to a certain MMR straight away. Regards, -E | ||
|
Excalibur_Z
United States12235 Posts
On November 19 2010 05:15 KaienFEMC wrote: I only have one question. In Taiwan server, the bottom top 200 players usually are people with very few games due to the small player pool. There are even platinum players who made it to the top 200 list (No. 198 30 wins with 4 loses). How would you explain this from your analysis/algorithm? Okay I just did a quick run-through of the top and bottom of the Taiwanese Top 200 (the top to get an idea of the S-class divisions and the bottom to see if there were any members of those divisions surrounding the Platinum guy, which there were). Here were the results: Blizz Name Record Rating Snapshot Division 1 MrGxASUS 446-216 2453 薩爾加斯 天微星 2 迅游高尔 582-241 2417 吞噬蟲 IOTA 3 NemoRanK 695-487 2526 腐化飛蟲 PHI 4 NvGoMax 130-38 2323 吞噬蟲 IOTA 5 OrzRush 340-170 2403 守衛者 天閒星 6 DusRush 358-201 2263 守衛者 天閒星 ... 195 Start 53-16 685 烏銳扎 天孤星 196 Gale 35-15 620 薩爾加斯 天微星 197 StarWbisu 46-18 579 守衛者 天閒星 198 袁鹏 30-4 1117 奧古斯格勒 天富星 199 吉姆莫里森 41-17 466 吉姆莫里森 200 超超 32-11 435 薩爾加斯 天微星 So we can see that the #198 guy, who is in Platinum, is between 620 and 435 points of division 薩爾加斯 天微星 which I presume to be S-Rank. Because there are no anomalies in points in the bottom 6, I believe the divisions of the #199 and #197 are also S-class, meaning the Plat guy is between 579 and 466. So 1117 in that guy's particular Plat division puts him between 579 and 466 S-Rank diamond. | ||
|
dGretch
United States28 Posts
so this system hinders them at both ends because the player will initially be placed into a lower class division, and then has no way to work out of it. even if divisions could be promoted from say A-class to S-class, then a very good player is still at the mercy of the others in his/her division, ¿no? it doesn't make any sense to me. let's say I quit my job and started playing sc2 10 hours a day non stop. even if i make marked improvements in my game so much so that i can compete with S-class players, i am still stuck in the same division i was promoted to from back when i only played 3 or 4 games a night. TL;DR, devaluing divisions doesn't make sense in my eyes because it doesn't account for players who actually drastically improve their game over time. | ||
|
Eeryck
United States184 Posts
On November 19 2010 06:14 dGretch wrote: my problem with the explanation is that it doesn't account for the fact that there will inevitably be players out there that do actually improve their play over time from average to pro level. so this system hinders them at both ends because the player will initially be placed into a lower class division, and then has no way to work out of it. even if divisions could be promoted from say A-class to S-class, then a very good player is still at the mercy of the others in his/her division, ¿no? it doesn't make any sense to me. let's say I quit my job and started playing sc2 10 hours a day non stop. even if i make marked improvements in my game so much so that i can compete with S-class players, i am still stuck in the same division i was promoted to from back when i only played 3 or 4 games a night. TL;DR, devaluing divisions doesn't make sense in my eyes because it doesn't account for players who actually drastically improve their game over time. If you were to drastically improve and were already in a low ranked diamond division you would start a huge winning streak getting well ahead of those in your division. Clearly marking you as outclassing your division because of your skill up. If you were to become number 1 in the world in hidden MMR, you would need to be about ~300-400 points ahead of someone in a S-Class division. The following season, you would find yourself placed in a higher class division. Blizzard has bet on there not being a ton of rags to riches stories. When there is one, or looks like there is going to be one, that is most likely what delays their promotion to diamond, trying to put them in a division that more closely fits their skill via demotion. | ||
| ||
