|
Hey Excalibur, do you have any idea what's up with people's MMR at the top of the ladder? Everyone still seems to be favored against everyone, even though it's been months since the reset and people have played upwards of a thousand games already. I guess it just means that everyone's MMR is much lower than their displayed rating, but the disparity between everyone's MMR seems to be artificially small even though their displayed ratings are not.
For example, I play select or idra on ladder, and I'm "slightly favored" against them. So it's like -17 or so for a loss. As far as display rating goes, they're over 100 points higher than me. Then I play some random player who is over 100 points lower than me, and I'm still "slightly favored" against them, taking about -19 points for a loss. Why is this disparity so small? About 90% of my matches have me being "slightly favored," with about 4% me being "favored" and 1% me being "evenly matched."
This was certainly not the case last season, where if I hit select on ladder he was always "favored" against me, and I took like -3 for a loss. Back then I'd say about 30% of my matches had me being "favored," 40% me being "slightly favored," and 30% my opponent being evenly matched or favored... in other words a lot more balanced.
Now when I play ladder there's just a lot of luck involved in who you happen to meet and players' MMR just don't seem to accurately reflect their true skill.
|
United States12248 Posts
On May 25 2011 23:44 Anihc wrote:
Hey Excalibur, do you have any idea what's up with people's MMR at the top of the ladder? Everyone still seems to be favored against everyone, even though it's been months since the reset and people have played upwards of a thousand games already. I guess it just means that everyone's MMR is much lower than their displayed rating, but the disparity between everyone's MMR seems to be artificially small even though their displayed ratings are not.
For example, I play select or idra on ladder, and I'm "slightly favored" against them. So it's like -17 or so for a loss. As far as display rating goes, they're over 100 points higher than me. Then I play some random player who is over 100 points lower than me, and I'm still "slightly favored" against them, taking about -19 points for a loss. Why is this disparity so small? About 90% of my matches have me being "slightly favored," with about 4% me being "favored" and 1% me being "evenly matched."
This was certainly not the case last season, where if I hit select on ladder he was always "favored" against me, and I took like -3 for a loss. Back then I'd say about 30% of my matches had me being "favored," 40% me being "slightly favored," and 30% my opponent being evenly matched or favored... in other words a lot more balanced.
Now when I play ladder there's just a lot of luck involved in who you happen to meet and players' MMR just don't seem to accurately reflect their true skill.
We saw the same thing happen at the end of Season 1, after the leagues were locked, remember? Players who were once earning +12/-12 against other top-level players were instead earning +2/-22 because the ceiling had been lowered. A bunch of players said they weren't going to continue playing for the rest of Season 1 because it was too risky to play games only to have their points start trending downward.
Anyway, this is completely normal and it happens in WoW Arena too. Once you reach the top of the ladder, it becomes harder and harder to earn more points. We saw the same thing even around the middle of Season 1 but it wasn't as pronounced because only the very very top players experienced it. Your win ratio has to be much higher than 50% to maintain your rating at the top of the ladder, and everyone else at the top of the ladder is experiencing the same thing. That's how you find out who is the best of the best: whoever is capable of getting the most points despite this handicap.
EDIT: Also, another possible reason for lowering the ceiling toward the end of Season 1 (which was shortly before the debut of Grandmaster) was to make spending bonus pool even harder for Grandmaster players and therefore make them play a lot more games to stay in GM league.
|
On May 25 2011 23:44 Anihc wrote:
Hey Excalibur, do you have any idea what's up with people's MMR at the top of the ladder? Everyone still seems to be favored against everyone, even though it's been months since the reset and people have played upwards of a thousand games already. I guess it just means that everyone's MMR is much lower than their displayed rating, but the disparity between everyone's MMR seems to be artificially small even though their displayed ratings are not.
For example, I play select or idra on ladder, and I'm "slightly favored" against them. So it's like -17 or so for a loss. As far as display rating goes, they're over 100 points higher than me. Then I play some random player who is over 100 points lower than me, and I'm still "slightly favored" against them, taking about -19 points for a loss. Why is this disparity so small? About 90% of my matches have me being "slightly favored," with about 4% me being "favored" and 1% me being "evenly matched."
This was certainly not the case last season, where if I hit select on ladder he was always "favored" against me, and I took like -3 for a loss. Back then I'd say about 30% of my matches had me being "favored," 40% me being "slightly favored," and 30% my opponent being evenly matched or favored... in other words a lot more balanced.
Now when I play ladder there's just a lot of luck involved in who you happen to meet and players' MMR just don't seem to accurately reflect their true skill.
This change actually happened at the last week of the season 1.
Imagine that the MMR of the top1 was 3900, the MMR of top 2 3800, top 3 3740, top 50 3400 etc.
There was actually a huge gap, now it is something like this:
Top 1: 3200, top 2: 3198; top3: 3193; top 50 3120.
The difference in MMR now isn't as huge as before, even though the difference in skill is still quite high. That means that the top 1 will only have that much more points than top 2 if he maintains his high win ratio, otherwise they'd have basicly the same points.
Something like this.
Edit: ExcaliburZ > me
|
On May 26 2011 00:20 SDream wrote:
Imagine that the MMR of the top1 was 3900, the MMR of top 2 3800, top 3 3740, top 50 3400 etc.
There was actually a huge gap, now it is something like this:
Top 1: 3200, top 2: 3198; top3: 3193; top 50 3120.
The difference in MMR now isn't as huge as before, even though the difference in skill is still quite high.
Yes, I hope you can see from reading my initial post that that was already my understanding as well. I guess my question is, will this get any better over time? It's already been many many weeks. Currently skill disparity >>> MMR disparity. Will skill disparity ever = MMR disparity like it was prior to the end of season 1?
On May 26 2011 00:15 Excalibur_Z wrote:
Your win ratio has to be much higher than 50% to maintain your rating at the top of the ladder, and everyone else at the top of the ladder is experiencing the same thing. That's how you find out who is the best of the best: whoever is capable of getting the most points despite this handicap.
My complaint is this: imagine you are rank 50 on ladder, and your MMR accurately reflects your true skill, and you play opponents whose MMR accurately reflects their true skill as well. If you play the #1 ranked person on ladder 20 times, you will lose many points and end up with a much lower rating than had you played the #100 ranked person ladder 20 times. This should not happen - if people's MMR truly reflect their skill then regardless of who they play their ladder rating should not change, and indeed this was the case prior to the end of season 1.
|
United States12248 Posts
On May 26 2011 00:58 Anihc wrote:Show nested quote +On May 26 2011 00:20 SDream wrote:
Imagine that the MMR of the top1 was 3900, the MMR of top 2 3800, top 3 3740, top 50 3400 etc.
There was actually a huge gap, now it is something like this:
Top 1: 3200, top 2: 3198; top3: 3193; top 50 3120.
The difference in MMR now isn't as huge as before, even though the difference in skill is still quite high. Yes, I hope you can see from reading my initial post that that was already my understanding as well. I guess my question is, will this get any better over time? It's already been many many weeks. Currently skill disparity >>> MMR disparity. Will skill disparity ever = MMR disparity like it was prior to the end of season 1? Show nested quote +On May 26 2011 00:15 Excalibur_Z wrote:
Your win ratio has to be much higher than 50% to maintain your rating at the top of the ladder, and everyone else at the top of the ladder is experiencing the same thing. That's how you find out who is the best of the best: whoever is capable of getting the most points despite this handicap.
My complaint is this: imagine you are rank 50 on ladder, and your MMR accurately reflects your true skill, and you play opponents whose MMR accurately reflects their true skill as well. If you play the #1 ranked person on ladder 20 times, you will lose many points and end up with a much lower rating than had you played the #100 ranked person ladder 20 times. This should not happen - if people's MMR truly reflect their skill then regardless of who they play their ladder rating should not change, and indeed this was the case prior to the end of season 1.
Can you post your match history and how many points you and your opponent won and lost for each game recently, along with who your opponent was? Vanick put it pretty well but I want to illustrate it further: "at the top end of the ladder skill becomes more spread out, as a result MMRs also get spread out so spread out MMRs makes the actual even matches rarer so you see +12/-12 less frequently, though they still do occur"
|
There could also be an effect of grandmasters league and its huge point offset.
GM had the top 200 mmr's but as it stands now I'm sure GM league is more of a grouping of 'some' of the very best players so you could be playing players in masters league who are just as good as GM's players, or better. Because of the difference in leagues the points earned from the matches will look different than they do at other levels.
Obviously's Excals answer is correct, but this could potentially explain why you can't understand the points you are getting from different players you face.
|
@Excalibur_Z
I still have trouble with that +/-12 story. I wonder "why" 12. Could it be because of an ELO system calculating the ratings?
I mean, with Warcraft III new players (first 80 games? I forgot the exact number) have a weightfactor 25, later 15, even later 10.
As it turns out, in an exactly even match you'd score 12.5 points per win
Rnew = Rold + WeightFactor(Result - ExpectedResult)
25 (0 - 0,5) = -12.5 [LOSS]
25 (1 - 0,5) = 12.5 [WIN]
If they rounddown the numbers it makes perfect sense that it's 12.
But then the next question would be, are the highlevel players who are "even matched" getting less points per win? (explained by a lower weightfactor)
I don't know if that's the case, is it?
Considered that people discribe gaining 6 points for a win and losing 19 for a loss might be perfectly logical if there is a 200 Rating difference and the weigth is still 25.
What's your thought on that?
|
United States12248 Posts
On May 27 2011 02:30 Gulzt wrote:
@Excalibur_Z
I still have trouble with that +/-12 story. I wonder "why" 12. Could it be because of an ELO system calculating the ratings?
I mean, with Warcraft III new players (first 80 games? I forgot the exact number) have a weightfactor 25, later 15, even later 10.
As it turns out, in an exactly even match you'd score 12.5 points per win
Rnew = Rold + WeightFactor(Result - ExpectedResult)
25 (0 - 0,5) = -12.5 [LOSS]
25 (1 - 0,5) = 12.5 [WIN]
If they rounddown the numbers it makes perfect sense that it's 12.
But then the next question would be, are the highlevel players who are "even matched" getting less points per win? (explained by a lower weightfactor)
I don't know if that's the case, is it?
Considered that people discribe gaining 6 points for a win and losing 19 for a loss might be perfectly logical if there is a 200 Rating difference and the weigth is still 25.
What's your thought on that?
The amount of points you can earn or lose in a game ranges from 0 to ±24. Therefore, 12 is right in the middle =)
We're still trying to determine how many points' difference equates to points earned in a game (that is, equal MMR and equal points for both players should produce a +12/-12 game, but what is the variance that causes a +13 or a +14 game?). That may not be known for some time.
|
On May 27 2011 03:09 Excalibur_Z wrote:Show nested quote +On May 27 2011 02:30 Gulzt wrote:
@Excalibur_Z
I still have trouble with that +/-12 story. I wonder "why" 12. Could it be because of an ELO system calculating the ratings?
I mean, with Warcraft III new players (first 80 games? I forgot the exact number) have a weightfactor 25, later 15, even later 10.
As it turns out, in an exactly even match you'd score 12.5 points per win
Rnew = Rold + WeightFactor(Result - ExpectedResult)
25 (0 - 0,5) = -12.5 [LOSS]
25 (1 - 0,5) = 12.5 [WIN]
If they rounddown the numbers it makes perfect sense that it's 12.
But then the next question would be, are the highlevel players who are "even matched" getting less points per win? (explained by a lower weightfactor)
I don't know if that's the case, is it?
Considered that people discribe gaining 6 points for a win and losing 19 for a loss might be perfectly logical if there is a 200 Rating difference and the weigth is still 25.
What's your thought on that?
The amount of points you can earn or lose in a game ranges from 0 to ±24. Therefore, 12 is right in the middle =) We're still trying to determine how many points' difference equates to points earned in a game (that is, equal MMR and equal points for both players should produce a +12/-12 game, but what is the variance that causes a +13 or a +14 game?). That may not be known for some time.
Well that supports my assumption that it's an ELO formula right? +12 is a rounddown of "12.5". in the ELO, max. weightfactor 25 (as was with warcraft III). whenever you have a winchance of 99,999% based on the difference in Rating the result is:
25*(0-.99) = -24.75 (rounddown = -24) LOSS
25*(1-.99) = 0.25 (rounddown = 0) WIN
I'd say it makes it's very likely that the visual rating system is a basic ELO.
|
It's possible but in my opinion unlikely. Since the system is still so opaque it can't be said one way or the other, but the selection of numbers does not necessarily mean anything, since those size numbers are already familiar to many players. Think of it this way: why use a totally different rating modification system for display points when Blizzard already has their new SC2 rating modification equations?
|
@Vanick
They already use two systems. If the visual ratingsystem is just a sugar sweet to keep us happy, why not use a basic form of ELO? They have plenty of experience with it (all blizz-online games before starcraft II). And isn't ELO the core of all the new systems that evolved including SC2s hidden rating system?
I'd say, if it's not ELO based, then would it mean they invented something completely different, and why?
The issue remains though, if it's that simple, then if player 1 would gain +12, player two would lose -12.. and that's not always the case.. still I can't shake the idea that it's somehow has to do with ELO, the -24 borders and +12 average is too much of a coincidence.. but the truth is somewhere close, something to do with comparing adjusted points and MMR as Excalibur explained.
|
Hi im playing top 50 GM EU and ive noticed that if top gm players play each other, theyre both slighty favoured respectively. if i win i get +9 and my opp. loses -16, if i lose i lose -16 and my opponent gets +9.
against high masters who play on the same ELO as gm's i lose -16 while he/she gets +12 (+/- 2 points)
is there a system installed which prevents us gm's from getting too many points?
in s1 there were ppl with more than 4000 points, now the best has barely 1200 after hundreds of games
|
On May 28 2011 00:00 DBS wrote:
in s1 there were ppl with more than 4000 points, now the best has barely 1200 after hundreds of games
Two reasons that contribute to that.
1) Bonus pool. Season 1 lasted about 8 months, and Season 2 lasted about 2 months so far. That's approximately 2300 extra bonus pool available.
2) Season 1 didn't have GM, and there is a points offset between GM and Master. In the Divison Tiers thread, best estimate of the offset seems to be about 430 pts.
4124 (Idra's score at end of S1) - 1238 (his current score) = 2886
(I picked Idra for no other reason than he is currently #1 in NA GM)
Bonus pool + GM offset
2300 + 430 = 2730
Not saying you're wrong about the points loss amounts being part of the reason too, I have no idea. Just saying, the bonus pool and offset do account for the vast majority of the discrepancy.
|
Ever since the ladder unlock and that I was promoted to diamond, some weird stuff has been happening to me. Every single player that I face, I am considered "Slightly Favored", losing 18 or 19 points a game, and winning 5 or 6 points a game. The thing is: all of them were diamond players with a reasonable amount of points, and ingame-skill level VERY similar to mine. all of them played plenty of games, so the system shouldn't have any doubt about our skill levels...right?
Can anyone explain to me why is this happening? There is nothing worse than losing 4 games for one loss, and this doesn't seem to fix itself over time - at least not yet.
Also, players that I beat lose 12 points instead of losing something like 10 or 8, or whatever the correct number would be.
|
On May 28 2011 05:59 Zephirdd wrote:
Ever since the ladder unlock and that I was promoted to diamond, some weird stuff has been happening to me. Every single player that I face, I am considered "Slightly Favored", losing 18 or 19 points a game, and winning 5 or 6 points a game. The thing is: all of them were diamond players with a reasonable amount of points, and ingame-skill level VERY similar to mine. all of them played plenty of games, so the system shouldn't have any doubt about our skill levels...right?
Can anyone explain to me why is this happening? There is nothing worse than losing 4 games for one loss, and this doesn't seem to fix itself over time - at least not yet.
Also, players that I beat lose 12 points instead of losing something like 10 or 8, or whatever the correct number would be.
If you lose 18-19, they should lose 7-6 in a "makes sense" logic.
It seems that for some reason they are in diamond rank E, while you're probably in a diamond rank C or B (just a guess).
Let's say you were promoted to a rank E, you'd have 73 adjustes points, but you would actually have a limit of -63 adjusted points to still be in diamond (because rank F exists).
But if you are in Rank C, then you will still start with 73 points (that are actually worth 199 points for a Rank E) and you can go as low as -189 adjusted points and still be in diamond "level".
So, if there was a bug, that bug was the one that let you go into a higher diamond rank, you shouldn't worry about it being "unfair" or "bugged" or something, ok? Good luck, have fun and congratulations on your high rank diamond :D
Edit: it's not impossible that there was a bug in your promotion and you are in a S rank diamond. But anyway, your mmr and points are working, we'd just have to figure out you division offset.
Edit:
Also, you are not alone:
http://www.teamliquid.net/forum/viewmessage.php?topic_id=227359
It confirms to be a bug and your new division is most likelly S Rank indeed, as I spotted a weird high number of latin american players getting promoted to our only one S Rank diamond division, that can't be a coincidence, for some unknown reason, players are being promoted to S Rank diamond divisions after the lock. As you guys weren't promoted as soon as you should have, the system got a little confused and promoted you to the highest tier (S Rank). That bug should, then, never happen again, and you guys will have to deal with it or demoted yourselves back to platinum and try to be promoted again into the "right" tier this time.
|
i have a question.
why is it when i had a 5-6 game winning streak, i lose one game and i get bumped down to bronze from silver?
I was at low silver for awhile but i constantly kept my rank moving up, almost to top 50. (unfortunately i dropped from a losing streak back to where i started, rank 70) and then i got another small winning streak up until some weird freak accident cause me to drop to bronze!?!?!
someone explain please? my mmr should be fairly balanced. i had a fair amount of losses when i started but i really fixed it with alot of wins. (bronze players are really not that good)
|
On May 28 2011 14:20 xmikeyy17x wrote:
i have a question.
why is it when i had a 5-6 game winning streak, i lose one game and i get bumped down to bronze from silver?
I was at low silver for awhile but i constantly kept my rank moving up, almost to top 50. (unfortunately i dropped from a losing streak back to where i started, rank 70) and then i got another small winning streak up until some weird freak accident cause me to drop to bronze!?!?!
someone explain please? my mmr should be fairly balanced. i had a fair amount of losses when i started but i really fixed it with alot of wins. (bronze players are really not that good)
There's no point winning against mediocre bronzes. If you wanted to stay silver, you'd have to win against most bronzes and also win against a decent amount of silvers.
A win is not always the same. A win against a bronze means way less than a win against a silver. If you win against 5 bronzes in a row, that doesn't matter, really. So, who did you lose to?
|
i checked back on it.
it was a bronze.
looks like i have to win ton of bronze games for little progress...
UUGHHHH
|
On May 27 2011 16:23 Gulzt wrote:
@Vanick
They already use two systems. If the visual ratingsystem is just a sugar sweet to keep us happy, why not use a basic form of ELO? They have plenty of experience with it (all blizz-online games before starcraft II). And isn't ELO the core of all the new systems that evolved including SC2s hidden rating system?
I'd say, if it's not ELO based, then would it mean they invented something completely different, and why?
The issue remains though, if it's that simple, then if player 1 would gain +12, player two would lose -12.. and that's not always the case.. still I can't shake the idea that it's somehow has to do with ELO, the -24 borders and +12 average is too much of a coincidence.. but the truth is somewhere close, something to do with comparing adjusted points and MMR as Excalibur explained.
I am fairly certain that the points system is a mostly generic ELO system that works on top of the more complex (yet ELO based) match making and promotion system.
The points system uses a constant 24 point modifier, where as the hidden MMR probably is using a scaling modifier based on the number of games you have played, and potentially your level of activity.
The reason points don't always matchup is because the points you earn are based off a comparison of your points to their mmr, and the points they earn is based off your mmr and their points. This means that two different calculations are used to determine points, thus the points will not add up.
|
I'm begining to think that in random team games the changes made to your mmr post match have to do not only with the level of skill of your opponents but also your relative skill on your team.
I say this because I was playing random 3's with my 2v2 teammate again last night. For the most part he has only played random 3's with me on his team (like 98%), while I had about 30 truely random games before I started playing with him. Without him I was a gold leaguer and he placed into platinum. As we continued to play he was promoted to diamond and I staid in gold. Then season two came and I placed into platinum while he placed back into diamond. We've continued to play and last night he was promoted to masters league.
Perhaps I have a diamond promotion looming at some point but for the most part we continue to play random diamond players (and arranged diamond teams occationally). It feels like he is getting a lot of credit for winning with me on his team while I am getting less credit because of him being on my team. I find it hard to believe that my early gold level games are holding me back so much, and I can't imagine my friend getting promoted to masters league without recieving some sort of handicap for me because we have yet to play a fully masters random team.
|
|
|
|
|
|
EPT
