Subscribe
Approximate Elos for new S13 rankings
Last edited on 19 July 2016 - 02:02 PM by Ny_
So after taking a small sample of the Elo ratings of various people in S12 build, I plotted them to see the approximate pattern of Elo. Gberry posted something similar in another thread, but for those curious:
Now, GBerry gave us approximate #'s for what percent of players should end up in each tier at the end of a season. The Elo from last season won't perfectly line up with the rating from next season, but theoretically, someone who is better than 90% of people in S12 will still be better than 90% of people in S13, so we should be able to compare the numbers somewhat.
Here's the estimates for the Build UHC ladder:
Masters: 99.99995th percentile | 2850+
Diamond: 99th percentile | 2000-2850
Emerald: 95th percentile | 1750-2000
Gold: 81st percentile | 1470-1750
Silver: 39th percentile | 1375-1470
Bronze: 0th percentile | 1375-
It's quite a bit lower than expected, but it makes sense because there aren't all that many 2k+ players, so the diamond rank has to be pretty spread out. Masters will be incredibly hard to get, even Danteh this season might still be a little bit lower than the cutoff.
If your ranking in the beta didn't match these, remember, there were a lot better players playing in the beta than usual, and build was often full of 2k+ players, so your ratings will be a lot worse than they would be in an actual season. Also, playing more should probably still reward you somewhat; grinding is less important, but will still help you, so playing just 20 matches might not be enough to show your true skill.
Posted on 17 July 2016 - 12:32 AM
Ny_ wrote
So after taking a small sample of the Elo ratings of various people in S12 build, I plotted them to see the approximate pattern of Elo. Gberry posted something similar in another thread, but for those curious:
Now, GBerry gave us approximate #'s for what percent of players should end up in each tier at the end of a season. The Elo from last season won't perfectly line up with the rating from next season, but theoretically, someone who is better than 90% of people in S12 will still be better than 90% of people in S13, so we should be able to compare the numbers somewhat.
Here's the estimates for the Build UHC ladder:
Masters: 99.99995th percentile | 2850+
Diamond: 99th percentile | 2000-2850
Emerald: 95th percentile | 1750-2000
Gold: 81st percentile | 1470-1750
Silver: 39th percentile | 1375-1470
Bronze: 0th percentile | 1375-
It's quite a bit lower than expected, but it makes sense because there aren't all that many 2k+ players, so the diamond rank has to be pretty spread out. Masters will be incredibly hard to get, even Danteh this season might still be a little bit lower than the cutoff.
If your ranking in the beta didn't match these, remember, there were a lot better players playing in the beta than usual, and build was often full of 2k+ players, so your ratings will be a lot worse than they would be in an actual season. Also, playing more should probably still reward you somewhat; grinding is less important, but will still help you, so playing just 20 matches might not be enough to show your true skill.
Now, GBerry gave us approximate #'s for what percent of players should end up in each tier at the end of a season. The Elo from last season won't perfectly line up with the rating from next season, but theoretically, someone who is better than 90% of people in S12 will still be better than 90% of people in S13, so we should be able to compare the numbers somewhat.
Here's the estimates for the Build UHC ladder:
Masters: 99.99995th percentile | 2850+
Diamond: 99th percentile | 2000-2850
Emerald: 95th percentile | 1750-2000
Gold: 81st percentile | 1470-1750
Silver: 39th percentile | 1375-1470
Bronze: 0th percentile | 1375-
It's quite a bit lower than expected, but it makes sense because there aren't all that many 2k+ players, so the diamond rank has to be pretty spread out. Masters will be incredibly hard to get, even Danteh this season might still be a little bit lower than the cutoff.
If your ranking in the beta didn't match these, remember, there were a lot better players playing in the beta than usual, and build was often full of 2k+ players, so your ratings will be a lot worse than they would be in an actual season. Also, playing more should probably still reward you somewhat; grinding is less important, but will still help you, so playing just 20 matches might not be enough to show your true skill.
Posted on 17 July 2016 - 01:00 AM
xToolBoxHD wrote
nice thread. But how exactly did you get these calculations?
So the percentages came from Gberry in this post. If we know Diamond is 1% and Emerald is 4%, then we know that anyone with Emerald is on at least the 95th percentile…they are in the top 5%.
With these percentages, we can look at last season's build Elo to find out how last season would translate to next season. As I said, I made a google spreadsheet and plotted around 50 random people's build elo and ranking on the leaderboard, and then used the ranking to figure out what percentile they are on. Here's the formula I used to find people's percentiles:
100 - [(Ranking / Total # of Players) * 100]
So for Danteh, we know he is #1. From what I could find, it looks like Build last season had 460,116 players. So if we sub into the formula:
100 - [(1 / 460116) * 100]
100 - .000217
99.999783
So Danteh is on the 99.999783rd percentile… or, Danteh is in the top .000217% of players.
With this formula, I can figure out the percentile of anyone in the ladder, which I did for 50 people. Then I put the numbers into the chart (which you can see in the main post).
Then, I estimated using MasterGBerry's numbers to figure out the conversion. If the top .00005% of players is Masters, and Danteh at 2800 is a bit below that, I have to assume that Masters would be a bit better than 2800. If Emerald is the top 5%, or the 95th percentile, and I see that a 1680 is on the 93rd percentile while a 1770 is on the 96th percentile, I estimated that the cutoff for the top 5% is about 1750.
The method isn't a hundred percent accurate, but the top 5% most skilled in season 12 will be similar to the top 5% most skilled in season 13, so we can compare the rating systems.
Posted on 17 July 2016 - 01:15 AM
99.99995th percentile is too high i think – that means you would have to be in front of 2 million players in order to enter mastersPosted on 17 July 2016 - 01:18 AM
Morsinius wrote
99.99995th percentile is too high i think – that means you would have to be in front of 2 million players in order to enter masters
Posted on 17 July 2016 - 01:18 AM
btw do you have a table for that chart you made? i'm curious to see the actual values you gotPosted on 17 July 2016 - 02:05 AM
I agree that Masters should have a higher proportion of players, the top 20-25 (~0.005%) perhaps?Posted on 17 July 2016 - 02:06 AM
Morsinius wrote
btw do you have a table for that chart you made? i'm curious to see the actual values you got
Posted on 17 July 2016 - 02:22 PM
Ny_ wrote
xToolBoxHD wrote...
So the percentages came from Gberry in this post. If we know Diamond is 1% and Emerald is 4%, then we know that anyone with Emerald is on at least the 95th percentile…they are in the top 5%.
With these percentages, we can look at last season's build Elo to find out how last season would translate to next season. As I said, I made a google spreadsheet and plotted around 50 random people's build elo and ranking on the leaderboard, and then used the ranking to figure out what percentile they are on. Here's the formula I used to find people's percentiles:
100 - [(Ranking / Total # of Players) * 100]
So for Danteh, we know he is #1. From what I could find, it looks like Build last season had 460,116 players. So if we sub into the formula:
100 - [(1 / 460116) * 100]
100 - .000217
99.999783
So Danteh is on the 99.999783rd percentile… or, Danteh is in the top .000217% of players.
With this formula, I can figure out the percentile of anyone in the ladder, which I did for 50 people. Then I put the numbers into the chart (which you can see in the main post).
Then, I estimated using MasterGBerry's numbers to figure out the conversion. If the top .00005% of players is Masters, and Danteh at 2800 is a bit below that, I have to assume that Masters would be a bit better than 2800. If Emerald is the top 5%, or the 95th percentile, and I see that a 1680 is on the 93rd percentile while a 1770 is on the 96th percentile, I estimated that the cutoff for the top 5% is about 1750.
The method isn't a hundred percent accurate, but the top 5% most skilled in season 12 will be similar to the top 5% most skilled in season 13, so we can compare the rating systems.
ok thank you
wow you must have a math degree :P
Posted on 17 July 2016 - 10:54 PM
You fucking smart ass <3Posted on 18 July 2016 - 02:27 AM
Ny_ wrote
Morsinius wrote...
whew yeah i was wondering how you'd gotten the values since it's only 20 pages now
very impressed you had the patience to go through it all and thanks, fascinating data
Posted on 18 July 2016 - 04:10 AM
Ur good at mathLast edited on 18 July 2016 - 04:18 AM by *deleted-41673
Automatically DeletedPosted on 18 July 2016 - 10:21 AM
Itaav91112 wrote
Ny_ wrote...
Sorry, but you made a big error here. Danteh is the best and that makes him top .000217% as you say. If the amount of players is the same. Even if Danteh would have got 3k. He still is the best and the amount of players is the same. He is still .000217 and even 3000 would still be Diamond.
To make it a bit more clear, because my Englsih isn't that good. With 2750 he would have been Build player #1. So if he had 2750 build you would say it is a bit better than 2750 instead of a bit better than 2800 and he would have Masters with 2800. With the same amount of players nobody will get Masters, because nobody can reach top .00005%. Or only the first person, but ONLY the first person would get masters since you can look at it that way
For everything else, nice you calculated this and it is really handy. Although it is possible the percentages is for the people who play often. High elos play on average more often so the average person ONLINE always had over 1400 ELO. If they calculate it that way a 1470 will NEVER have Gold. Just Iron IV or something. This would be less logical but easier to calculate
EDIT: My second point is not really important, but if you don't understand the first part because my English is pretty terrible, I wouldn't mind explaining it again
Posted on 18 July 2016 - 01:22 PM
Ny_ wrote
Gold: 81st percentile | 1470-1750
Posted on 19 July 2016 - 02:04 PM
Khazars wrote
Ny_ wrote...
Last edited on 19 July 2016 - 02:08 PM by crna_glava
Ny_ wrote
Khazars wrote...
I am Emerald V now btw i think it should be like this:
Bronze: Unknown - 1300
Silver: 1300 - 1600
Gold: 1600 - 2000
Emerald: 2000 - Unknown (1 guy who was 2400 last season is now Emerald IV lol)
Diamond: ????
Master: ????
Posted on 19 July 2016 - 03:11 PM
Ik I'm wrong and I have no math to back this up but this is what makes sense to me.Brown: Under 1300
Silver: 1300-1600
Gold: 1600-1900
Emerald: 1900-2200
Diamond: 2200-2500
Masters: 2500+
?
Posted on 19 July 2016 - 08:12 PM
I think you guys are looking at this and assuming the numbers should be higher because 2k+ people are being placed in gold/emerald, but I have to remind you, think about how ELO goes over a season. In the beginning, you have way fewer players and there's way less ELO available, so it's not really spread out. Remember, on the first day of a new season, the top players are usually under 2k in every ladder. So a top tier player being placed in low Emerald (~1800), makes sense, especially since you're forced to play better players because of the new matchmaking. I imagine the matchmaking also probably keeps ELO from being as inflated as it usually gets.Anyway, I got Emerald V in archer after going 16-4 against a lot of good players, which is around 1750…I expect it to go up to Emerald I-II as the season progresses, which would reflect my average 1900-2000 rating from past seasons. The low level comparisons seem pretty accurate though, I have Emerald V in archer (1750ish) and haven't played anything else, which would be a global of about 1430…and my global is Silver II, exactly what you'd expect.
Posted on 19 July 2016 - 08:38 PM
Wouldn't this only be accurate when there is 460,000 players in the ladder?