|
On November 02 2010 03:39 darmousseh wrote:Show nested quote +On November 02 2010 03:29 Mendelfist wrote: As I already said, your formula doesn't seem to work at all if I look at my own match history.
Some examples: Match 1: Mendelfist, Silver, 753 points. MMR = (1 - 1) * 500 + 753 + 317 + 1500 - 862 = 1708 Wilu, Silver, 982 points. MMR = (1 - 1) * 500 + 982 + 153 + 1500 - 862 = 1773
Match 2: Mendelfist, Silver, 779 points. MMR = (1 - 1) * 500 + 779 + 304 + 1500 - 862 = 1721 Nemesis, Gold, 743 points. MMR = (2 - 1) * 500 + 743 + 160 + 1500 - 862 = 2041
Match 3: Mendelfist, Silver, 769 points. MMR = (1 - 1) * 500 + 769 + 305 + 1500 - 863 = 1711 Zbooj, Silver, 665 points. MMR = (1 - 1) * 500 + 665 + 376 + 1500 - 863 = 1678
Match 4: Mendelfist, Silver, 821 points. MMR = (1 - 1) * 500 + 821 + 318 + 1500 - 902 = 1737 BBang, Gold 855 points. MMR = (2 - 1) * 500 + 855 + 109 + 1500 - 902 = 2062
Within a league the calculated MMRs are close, but across the leagues they are way too different. The matches were called even by the matchmaking system, they felt even so the MMRs should be close. Your formula does not work. Either the 500 point offset is wrong or there are other pieces of the puzzle missing.
Like i said, the formula doesn't work too well for streaky players. MMR can change by a lot more than points can during a time. The formula is more for solidified players than players who are currently improving. How quickly were you promoted after these matches? And did you play a lot of favored games once you were promoted?
There were no streaks involved on my part here.
The 30 matches before this were 15 wins and 15 losses. Longest win streak: 3 matches. Longest loss streak: 3 matches. All 30 were "even". My opponents in my four example matches had 141, 151, 67 and 158 games played, so they should be fairly stable. I won both gold matches and lost one of the silver matches.
I played 47 matches after "match 4" before I was promoted to gold. I have not been favored in any matches after the example matches. A vast majority of them were "even".
Do you see why I think there is something wrong with your formula?
|
Here is something interesting Idra US (not as active) vs EGIdra KR (active) MMR: 3733 vs 3956 LR: 2219 vs 2278 Games: 168/47 vs218/132
so it seems that how idra, artosis, etc were always saying that US was much weaker than KR, it's just that there are more players overall watering it down, making it seem weaker, but in reality it's about the same. Idra just played more games with lesser skilled players as he was below his MMR gaining him a much better win % and much weaker opponents.
|
I'm creating all the top zerg players from korea right now so I can see who the best zergs are at the moment. But some of the names are in korean, how do I add those? And does anyone know who is who?
|
Can you see when you are getting promoted by this? Or isn't it known at wich MMR you get promoted?
|
On November 02 2010 02:49 darmousseh wrote: So i thought, what if blizzard is being retarded and simply offsetting the divisions by some points? So i decided to go with 500 for testing purposes. Division offset * 500 + points. I took some random data from all of the divisions, applied the formula and checked the distribution and guess what, it was perfect (for those with at least 200 games).
At this point I am very confident, if you have used your entire bonus pool it is easy to compare yourself to other divisions. Then i assumed that what if i just added the bonus pool remaining for those who haven't completed the games. So i added the bonus pool (estimation since i don't actually have the data. the estimation was 12 * wins used) and added them to the grouping and the distribution was still equal except for people under 50 games.
Actually what you did was remove the effects of the bonus pool, which makes sense because the bonus pool points substantially distort the visible ratings.
[ Visible Points ] + [ Bonus Remaining] + [ 1500 - Total Bonus ]
=
[ Base Points ] + [ Bonus Received ] + [ Bonus Remaining ] + [ 1500 - Total Bonus ]
=
[ Base Points ] + [ 1500 ]
According to your formula, I as a 1022 points Silver league, with 415 bonus points remaining, and 1340 total bonus pool (taken directly from my 2v2 ladder, with 0 wins after my placement games, so should reflect total bonus pool), my rating is [ 500 * (1 - 1) ] + [ 1022 ] + [ 415 ] + [ 1500 ] - [ 1340 ] = 1597 rating. A grand total of 97 of those points (6.1%) comes from my actual skill. Put another way, I have 1022 visible points, 925 of which come from the bonus pool.
I think that applying the division adjustment makes no logical sense. If I was playing the exact same opponents, getting the exact same results, but was placed in Bronze League, I would have a whopping 500 points less, just because of the division I was in. That makes no logical sense if the point of the MMR is to match me against people of equal skill. None at all.
If that's how the points actually work, then in my view it's a fundamentally flawed system. It would, however, help explain why people get stuck in Bronze. Playing and beating platinum level players, they would be at a 1000 point disadvantage compared to Gold League players who were doing the same thing.
|
I just thought of a better way to verify if this is correct. I am going to create a simulation of bnet ladder assuming that they are using this system and check the results. I think I will do that tonight. I will provide the source code whether or not i prove myself correct or false. The way i will test my hypothesis is if the distribution in my simulation is exactly the same as the distribution of bnet.
Also 2v2's are different than 1v1s. Your 2v2 rating for a match is dependent on both players ratings combined.
|
Listen darmousseh, I really appreciate your hard work on this topic. But you are really screwing something up here. The new US Top 200 have just been posted. If you compare them to sc2ranks.com, you will definitely see, that the divisions are a far too important factor to be ignored.
Besides that, half of the posters here in this thread even ignore the fact, that your site does not know about the bonus points consumed (as is sc2ranks.com).
Here a snapshot of sc2ranks (9PM GMT) compared to newest top 200 list from nov 1st.
<sc2ranks> <player> <top 200 rank> 1 AlLaboUtyOu 3 2 SeleCT 1 3 dayvie 43! 4 HuK 2 5 DdoRo 38! 6 ACERGAME 102!! 7 scvrusher 60! 8 Valiant 63! 9 RuFF 94!! 10 theognis 22
*edit: quite a lot of typos, sry*
|
On November 02 2010 05:31 Merano wrote:Listen darmousseh, I really appreciate your hard work on this topic. But you are really screwing something up here. The new US Top 200 have just been posted. If you compare them to sc2ranks.com, you will definitely see, that the divisions are a far too important factor to be ignored. Besides that, half of the posters here in this thread even ignore the fact, that your site does not know about the bonus points consumed (as is sc2ranks.com). Here a snapshot of sc2ranks (9PM GMT) compared to newest top 200 list from nov 1st. <sc2ranks> <player> <top 200 rank> 1 AlLaboUtyOu 3 2 SeleCT 1 3 dayvie 43! 4 HuK 2 5 DdoRo 38! 6 ACERGAME 102!! 7 scvrusher 60! 8 Valiant 63! 9 RuFF 94!! 10 theognis 22 *edit: quite a lot of typos, sry*
I'm beginning to think that the top 200 takes into account recent activity more significantly than overall points.
|
This is fucking retarded (not the thread, but the system). Blizzard needs to just let us know our fucking ratings instead of forcing us to data-mine and try to deduce their mysterious system.
Anyone who went to Blizzcon know anything about the ETA of their next big "esports patch?" I thought I heard something about them making this system more public.
|
On November 02 2010 06:05 SpaceYeti wrote: This is fucking retarded (not the thread, but the system). Blizzard needs to just let us know our fucking ratings instead of forcing us to data-mine and try to deduce their mysterious system.
Anyone who went to Blizzcon know anything about the ETA of their next big "esports patch?" I thought I heard something about them making this system more public.
I agree with you 100%. I'm a platinum player and I could care less about what division i'm in. Maybe if divisions had more meaning then i would care. One thought I had was to seperate divisions by location. Being the top bay area platinum player has a whole lot more meaning than the top platinum player of 100 random people.
In any case you should be able to see your MMR (mean and sigma). People who play 1v1 ladder aren't 10 year old retards. Even xbox live lets you know if you suck. They could make it easy. They should display ratings as confidence intervals.
So if you are 1500 mean, 150 sigma(a typical starting player). Then it should display your rating as 1500(1200-1800) and later on if you are 1600 mean, 25 sigma. Then you will be 1600(1550-1650)
This is very simple to understand and is very true.
The only problem with this system is that it might discourage people to continue playing since they won't be earning any rating once they reach a peak. Chess has had this problem for years, but they gave up and figured people would keep playing because they wanted to play chess, not because they wanted to increase their rating.
|
On November 02 2010 06:31 darmousseh wrote: So if you are 1500 mean, 150 sigma(a typical starting player). Then it should display your rating as 1500(1200-1800) and later on if you are 1600 mean, 25 sigma. Then you will be 1600(1550-1650)
This is very simple to understand and is very true.
The only problem with this system is that it might discourage people to continue playing since they won't be earning any rating once they reach a peak. Chess has had this problem for years, but they gave up and figured people would keep playing because they wanted to play chess, not because they wanted to increase their rating.
I couldn't agree more with both points.
Displaying our actual MMR, or at least a separate-but-related non-inflationary rating (á la WoW arena), makes so much more sense than the awful Bonus Pool system if you're going to rank people based on it. If I'm going to be ranked above or below somebody, I want it to be because I'm better or worse than them. I don't want to be tricked into thinking I'm better just because I've played more.
Also, BW iCCup has shown that players are willing to continue playing a very similar game for over a decade while using a rating system. The idea that Blizz has to constantly inflate our ratings over the course of the season to keep us interested is, I think, a gross underestimation of the player base. It's bordering on insulting.
|
On November 02 2010 07:27 The Touch wrote:Show nested quote +On November 02 2010 06:31 darmousseh wrote: So if you are 1500 mean, 150 sigma(a typical starting player). Then it should display your rating as 1500(1200-1800) and later on if you are 1600 mean, 25 sigma. Then you will be 1600(1550-1650)
This is very simple to understand and is very true.
The only problem with this system is that it might discourage people to continue playing since they won't be earning any rating once they reach a peak. Chess has had this problem for years, but they gave up and figured people would keep playing because they wanted to play chess, not because they wanted to increase their rating. I couldn't agree more with both points. Displaying our actual MMR, or at least a separate-but-related non-inflationary rating (á la WoW arena), makes so much more sense than the awful Bonus Pool system if you're going to rank people based on it. If I'm going to be ranked above or below somebody, I want it to be because I'm better or worse than them. I don't want to be tricked into thinking I'm better just because I've played more. Also, BW iCCup has shown that players are willing to continue playing a very similar game for over a decade while using a rating system. The idea that Blizz has to constantly inflate our ratings over the course of the season to keep us interested is, I think, a gross underestimation of the player base. It's bordering on insulting.
Yeah. Yahoo games has a typical ELO system and people continue to play that regardless of their rating. If the game is good enough, we shouldn't need an incentive to play. Also the ladder has kinda drawn me away from playing. I'm not very likely to use up all my bonus points and without them it's impossible to tell how good i am.
|
On November 02 2010 06:05 SpaceYeti wrote: This is fucking retarded (not the thread, but the system). Blizzard needs to just let us know our fucking ratings instead of forcing us to data-mine and try to deduce their mysterious system.
Anyone who went to Blizzcon know anything about the ETA of their next big "esports patch?" I thought I heard something about them making this system more public.
there's supposed to be a minor patch soon I'd guess 8th or 9th (right after MLG Dallas) if not, then maybe 14th (right after GSL finals)
I think the big patch won't be for a few weeks at least (probably closer to new year)
and I agree that transparency is needed. All Blizz needs to do is release how they calculate MMR/trueskill/whatever they call it smart people/fans like Shadow, dahmoussey, excaliburz, etc. can run their own sites with correct rankings, and Blizz can keep their points system for all the casuals who couldn't care less
|
On November 02 2010 04:55 darmousseh wrote: I just thought of a better way to verify if this is correct. I am going to create a simulation of bnet ladder assuming that they are using this system and check the results. I think I will do that tonight. I will provide the source code whether or not i prove myself correct or false. The way i will test my hypothesis is if the distribution in my simulation is exactly the same as the distribution of bnet.
Also 2v2's are different than 1v1s. Your 2v2 rating for a match is dependent on both players ratings combined. Depends on the 2v2 type. Arranged teams have their own specific MMR. 2v2 Random uses the combination of the players ratings as you said.
Anyways, I'm not sure your simulation will show anything. All you're doing is offsetting the points that people already have by a common value, so of course the points will be distributed the same as SC2Ranks. There is nothing really changing. Basically, what your thread is actually about is the MMR offsets between leagues, for which you have offered no proof or valid argument. The main reason why the deduction you're attempting can't work is divisions within leagues tweak the points values. A lot. To the point where sc2ranks is very inaccurate except at the top200 area where it's just sort of inaccurate. This is not to put sc2ranks down as it still provides a very valuable service and Shadowed put a lot of work into it, but the caveat remains.
There can be a vast gap in MMRs between the #1 guy in platinum's best league, and the #1 guy in platinum's lowest division. This fact alone will stop your analysis from bearing fruit.
|
Hey thanks for trying to figure out how to calculate your mmr Am highly interested to see my own progress in skill but am unable to because blizzard somehow refuses to make the mmr public, understand why they do the ladder thing only but people who realy interested in this game are much more interested in the true skill indicator then in the ladder competition (at least this goes for me) Annyway beside this i have 1 question/remark According to this aproach each time you get promoted you should loose a totall of 500 points combined in both your bonus pool left and the ladder points you have (because your mmr doesnt change just because you get promoted) and am not sure i recall correctly but it seemed to me that i lost alot less then thoose 500 points the last time i was promoted This has been 2-3 months ago and i think i maybe lost 200 points or so in my ladder though i think i even gained in the bonus pool Not sure if other people have hard data about this but i realy doubt you loose 500 points combined on both your bonus pool and ladder pool when you are promoted
The idea that Blizz has to constantly inflate our ratings over the course of the season to keep us interested is, I think, a gross underestimation of the player base. It's bordering on insulting. Last edit: 2010-11-02 07:27:35
have to admit i am also realy disapointed in this and would think that if blizzard took esports serious that they would also aply a true skill indicator guess the vast majority of sc2 plays the game casual and is more interested to see progress on ladder then a real nr to messure skill still i see no harm in making the mmr visable for people who would like to see it It would also be alot easier to judge peoples skill lvl instead of the ladder points wich are horrible for that
|
5673 Posts
The OP has not responded to the arguments of known community ladder analysts and statisticians like Excalibur_Z, Vanick or Shadowed and blindly continues to expound his theory in the face of facts that prove him wrong. This thread is closed in the interest of containing misinformation. Please continue discussion in the original ladder analysis thread.
|
|
|
|