EPTAn incomplete guide to bnet 2.0 ranking system - Page 3
| Forum Index > SC2 General |
|
Sfydjklm
United States9218 Posts
| ||
|
Sfydjklm
United States9218 Posts
| ||
|
zomgzergrush
United States923 Posts
Would it really kill blizzard for someone to just post on their bnet forum or something about how the point system works? | ||
|
Tamerlane
Canada424 Posts
On March 16 2010 07:21 edahl wrote: You just need one formula y=12+x+z. I do suggest a slightly better notation letting p mean points, b mean bonus points and f mean points of favour. You might also want to let c be your constant 12, just because that is also subject to change. For brevity (and readability) your formula formula then reads p=c+b+f, with c=12, b in {0,...,k} (where k is your bonus pool), and f in {-12,...,12}. Slight favour in either direction means f in {-3,...,3}, strong favour means f in {-12,...,-4, 4,...12}. Your suggestions are definitely worthy, I'll formalize my first post a bit which I had not bothered doing in the first place. On March 16 2010 07:21 edahl wrote:EDIT: I conjecture something down the following line. Let K={k1 (practice), k2 (copper), ..., k7} be a set of constants with k1<...<k7 (this set might simply be K={1,...,7}). Let p,q be players of some match m with ratings r(p),r(q), and let k(p),k(q) in K be the values in K associated to the players p,q. Let w(m) be the winner of the match, and l(m) the loser. Let m(w(m))=1, m(l(m))=-1 (we may call this the outcome sign). Then there exists a function deciding f(x) with x in {p,q} (deciding the f of each player) given by SOMETHING LIKE f(x)=m(x)|r(p)-r(q)|/(|k(p)-k(q)|+1). Now, my proposed f is not the correct one, but it might get some of you down the correct line of reasoning. The functions I have provided may not be sufficient, i.e., there may be more cases and factors and relations cropping up. TBH, I'm not sure what is the meaning of f(x) and m(x), nor how we could ever determine the constants involved. On March 16 2010 16:26 Sfydjklm wrote: that would mean if i decide to play offrace lets say drop 300 pts or so and then go 20-10 against silver-gold guys instead of going 15-15 against platinum guys i get bumped into plat? I didn't suggest there was a "checkpoint" (or bump) every 30 games, I would be surprised this was the case (and in fact, I have been reranked at my 112th game) and I'm fairly sure your opponent's true/hidden rating is factored in, which means your record or win ratio is of little use to determine if and where you get promoted/demoted. | ||
|
edahl
Norway483 Posts
On March 16 2010 22:09 Tamerlane wrote: TBH, I'm not sure what is the meaning of f(x) and m(x), nor how we could ever determine the constants involved. m is a function from {p,q}->{-1,1}. If p is the winner, m(p)=1 and vice versa. f=f(x) is a function from {p,q}->{-12,...,12}, and is simply the function that determines your f after a match. If p wins 12 points from the opponent being very favoured, f(p)=12. Similarily, f(q) might be -12 in this situation. It's kinda like doing physics: At first you'll just have to guess it, then you have to try it out. There's no point in determining the exact constants just yet, but it makes sense that they're there. They may be based on some distribution, or they might increase linearly. | ||
|
Disastorm
United States922 Posts
On March 14 2010 19:32 Tamerlane wrote: |y| = 12 + |x| + z I know it seems like you put alot of work into this, but unfortunately the original poster's equation is wrong. At higher levels, the "x" becomes more negative for each matchup. For example, at higher ELO levels an even matchup will actually have a negative x, and a slightly favored opponent will still even have a negative x (although less negative than the even matchup). On March 14 2010 19:32 Tamerlane wrote: The value of x is exactly the reason why top players were often getting 0-2pts from wins and losing 20+ when they lost early on, their opponents were rated much lower. Well x is the reason, but actually its not because their opponents are rated much lower, its in fact because even at the same ratings, the x value is lower than with lower ELO. I'm not sure why this was done, it was most likely done to put a theoretical limit of rating points one can obtain, as perhaps they didn't want anyone with like 50,000 points. | ||
|
oxxo
988 Posts
| ||
|
Piousflea
United States259 Posts
Like "level" in Halo 3 or Warcraft 3, the "Rating" is not intended as a measure of player skill. It is a function of both player skill and the number of games played since it generally increases more for winning than it decreases for losing. However, there is a hidden "skill rating" that the game uses to track how good a player is. The skill rating is a zero-sum ELO-like score. I don't think that we will easily be able to determine how that rating is calculated. Also, we know that most modern matchmaking systems use an "uncertainty" factor as well. If you consistently win against worse players and lose against better players, the matchmaking system can very confidently say what your skill level is. On the other hand, if you alternate between losing against bad players and then winning against good players, it really does not know what your skill is. This is likely to be a factor in advancing leagues, since it doesn't want to advance (or demote) you if it is very uncertain about your true skill. | ||
|
Tamerlane
Canada424 Posts
On March 17 2010 03:07 Disastorm wrote:I know it seems like you put alot of work into this, but unfortunately the original poster's equation is wrong. At higher levels, the "x" becomes more negative for each matchup. For example, at higher ELO levels an even matchup will actually have a negative x, and a slightly favored opponent will still even have a negative x (although less negative than the even matchup). Well x is the reason, but actually its not because their opponents are rated much lower, its in fact because even at the same ratings, the x value is lower than with lower ELO. I'm not sure why this was done, it was most likely done to put a theoretical limit of rating points one can obtain, as perhaps they didn't want anyone with like 50,000 points. I think you missed (if you did read?) the part where I explain x isn't based on actual "ELO" (it's not ELO guys, stop taking about ELO, I know you're used to the ratings of SC1/WGT/iCCup, but please move on) but rather an expected rating. That means 3 things directly regarding your post : - it totally makes sense that x becomes more and more negative for player as he get at the highest level => that's because his expected rating his higher than most of the players he is facing - an "even matchup" is by a definition a matchup where no players is favored (at least, that's what the system believes), x won't be negative in that case (it will be equal to zero) - it's not done to put a theorical limit, it's a different system than ELO => if my assumptions on x are correct, it's probably similar to the glicko system (which uses ratings deviation instead of ratings difference, such as in ELO) On March 17 2010 03:32 oxxo wrote: I've gone silver->plat->silver->plat any idea on the number of games played at these switches? On March 17 2010 09:15 Piousflea wrote: My suspicion is that the skill rating system is similar to Xbox Live (Halo) rating. The "Trueskill" rating system of the XBox Live is based on the Glicko 2 rating system, it's even posted on their website. | ||
|
Dionyseus
United States2068 Posts
| ||
|
Tamerlane
Canada424 Posts
Only 2 games into this experiment and I got promoted to platinum, in a new division that had only 24 players in it...and that occured after 2 losses in a row. My starting rating was 1011 (down from ~1804 in silver) and after the 3rd game (which I lost as well) I could now get an update on the bonus pool and rested points I had : 464 and 107 respectively. Over the next few games and division growth, I could verify 3 important information : - We earn 1 rested point every 4 hours. - This is no initial period required to earn rested points. - We earn 1 bonus point for every player that joins the division. At least, that's the only explanation I could find between the evolution of my bonus pool and rested points as there was a few points discrepancy that equaled the number of players added to the division. Also, now that the 4 hours period was verified, I did a little calculation : => I was being reranked on March 15, with 107 rested points, divide that by 4 and it equals 26.75. 26 days prior to March 15 was February 17, the day the beta launched. Finally, on the favored indicator : my rating didn't change much during that time, staying in the 1100-1150 range for a while as I was trying a few things and lost consistently. Furthermore, ALL of my opponents had higher actual rating than I did, the only thing that changed was the league in which they played and since I was in a new division and my rank was always 3rd/4th, I was ranked higher than almost all of them (except a few guys in the bronze/silver/gold leagues). Here is the breakdown of my 22 recorded games as platinum when it comes to league+rank : Favored (including slightly) - 4 platinum players, ranking between 16 and 56 - 3 gold players, ranking between 7 and 15 Opponent is favored (including slightly) - 5 platinum players, ranking between 16 and 38 - 8 gold players, all ranked between 3 and 7 (except for 1 player who was 15) - 1 silver player, ranked 1 - 1 bronze player, ranked 1 Even matchup - 2 gold players, ranked 5 and 11 Conclusion (albeit a relatively small sample, I believe it is still conclusive) : - There is NO correlation between the actual rating and "favored" status - There is NO correlation between the rank of player and the "favored" status (and this totally makes sense as little divisions, where almost everyone is ranked higher than should be would be at a huge disadvantage when it comes to earning points) - Corollary of the previous point : there is NO correlation between the division size and the points earned (appart from the bonus pool). | ||
|
Icks
France186 Posts
| ||
|
onmach
United States1241 Posts
Also, now that the 4 hours period was verified, I did a little calculation : => I was being reranked on March 15, with 107 rested points, divide that by 4 and it equals 26.75. 26 days prior to March 15 was February 17, the day the beta launched. You should have divided by 6. You get 6 points per day, that would equal 17.8 days, which would put you at the end of february, which means nothing. By the way, did you keep track of how many game you had the moment you got promoted? It would be nice to know when to expect a second chance to be reranked. | ||
|
Tamerlane
Canada424 Posts
On March 18 2010 23:46 Icks wrote: Did you notice the number of games played by the "favored" bronze and silver players? Were they new players? If not, i dont see how they could be favored against a platinum... I didn't note the exact number of games, but that was above 100 for sure, their rating was 1829 and 1786 for bronze and silver guys respectively. On March 18 2010 23:56 onmach wrote: You should have divided by 6. You get 6 points per day, that would equal 17.8 days, which would put you at the end of february, which means nothing. ahhh...you're right, I don't know how I could have done such a simple mistake :O well, it means nothing except if we get rested points at a rate of 1 per 6 hours! => I'll check this out after the reset (I didn't note the exact time at which I got y rested points, but the 1pt/4h rate seemed to do fit - although it could have happened that I got 2pts in a lapse of anywhere between 8 and 12 hours even if the rate was 1pt/6h period!) On March 18 2010 23:56 onmach wrote:By the way, did you keep track of how many game you had the moment you got promoted? It would be nice to know when to expect a second chance to be reranked. I thought I had mention this tbh, it was after game #112 ...which seems kind of random. I'm starting to believe that reranking can occur at pretty much anytime after game 30! | ||
|
Icks
France186 Posts
On March 19 2010 01:10 Tamerlane wrote: I thought I had mention this tbh, it was after game #112 ...which seems kind of random. I'm starting to believe that reranking can occur at pretty much anytime after game 30! If the reranking period is not random, at least Blizzard will change the time and frequency regularly. They said in the FAQ that the time and frequency will be kept hidden. If it was the same time and frequency everytime, it would be easy for anyone to understand how it works, the way you're trying to do it. And it wouldn't be hidden for a long time. So it's certainly "random after a certain amount", as you said. | ||
|
gorchiza
Bulgaria200 Posts
one side question pls: how we can reset our stats ? | ||
|
Disastorm
United States922 Posts
On March 18 2010 02:14 Tamerlane wrote: I think you missed (if you did read?) the part where I explain x isn't based on actual "ELO" (it's not ELO guys, stop taking about ELO, I know you're used to the ratings of SC1/WGT/iCCup, but please move on) but rather an expected rating. That means 3 things directly regarding your post : - it totally makes sense that x becomes more and more negative for player as he get at the highest level => that's because his expected rating his higher than most of the players he is facing - an "even matchup" is by a definition a matchup where no players is favored (at least, that's what the system believes), x won't be negative in that case (it will be equal to zero) - it's not done to put a theorical limit, it's a different system than ELO => if my assumptions on x are correct, it's probably similar to the glicko system (which uses ratings deviation instead of ratings difference, such as in ELO) any idea on the number of games played at these switches? The "Trueskill" rating system of the XBox Live is based on the Glicko 2 rating system, it's even posted on their website. Ok srry i didnt know about the non-ELO system, however it doesn't change the fact that I have won games against slightly favored people and lost points due to "x". And no I am not referring to the "message bug" displayed in the result screen where it falsely says your opponent was favored. I am referring to the loading screen which blizzard has said is not bugged. However, thinking back on it, I don't think it happens that often and so it is possible its just a bug with the scoring system. | ||
|
Tamerlane
Canada424 Posts
On March 19 2010 17:42 Disastorm wrote:Ok srry i didnt know about the non-ELO system, however it doesn't change the fact that I have won games against slightly favored people and lost points due to "x". And no I am not referring to the "message bug" displayed in the result screen where it falsely says your opponent was favored. I am referring to the loading screen which blizzard has said is not bugged. However, thinking back on it, I don't think it happens that often and so it is possible its just a bug with the scoring system. Losing points after a win...that's unheard of!! If it happens again, could you please take a screenshot? I'll make sure that is reported as a bug (and if it isn't a bug, hopefully we'll get some more info on how the system works) @gorchiza : it's not possible to reset your stats, Blizzard made it clear they wanted to eradicate smurfing, although the beta ladder will be reset (well, all accounts will be deleted to be exact) in the near future | ||
|
Sfydjklm
United States9218 Posts
| ||
|
ManiacTheZealot
United States490 Posts
| ||
| ||
