|
On March 08 2011 08:14 Excalibur_Z wrote:
Yeah definitely, it has to calculate how many points the game will be worth before points are actually awarded, which means comparing MMRs from the beginning.
That gap is pretty huge, more than I would have expected. 29 point difference, sure okay. 71 point difference... I've gotta think that's a separate tier.
Edited: In the diamond examples the difference before match is only 5 points taking in consideration the high possibility of different tiers.
Edit7: Considering that players can "eat" their bonus points if they lose enough points without being demoted, than it is interesting to think that the math on some players will have this distortion, so how could we guarantee this has never happened? o.o
|
On March 08 2011 07:18 Excalibur_Z wrote:Show nested quote +On March 08 2011 06:20 Candles wrote:
Ok this just happened and I thought it might help. Just got promoted from:
Bronze (Broodling November) to Silver (Tosh Upsilon) My points dropped from 1949 (including the win that caused the promotion) to 1838. That seems to a drop of 111. You said multiples of 56 which would give 112. Maybe I've got the maths wrong, but I don't thing so... Whenever you change leagues, your points get reset to 73 + spent bonus pool. So that means you've spent 1765 bonus pool. How many points did you earn for your last game?
My last game (The one I got promoted on) gave me 24 points. (12 bonus + 12 normal) Strangely the game isn't showing on the opponents match history so i couldn't check if he also lost 12 points so I could give you the data.
You're maths seems very good so far and so I feel I'm gonna regret saying this, but there seems no way I've used up 1763 bonus pool. That would mean I'd gain a gross of 1763x2 = 3526 points in my career if i'd never emptied my pool, which I haven't
Based on the fact I've won approximately 50% of my matches and I always gain roughly 2x points for a win and lose x points for a loss (because of bonus pool) then my final points total would have been 2/3 x 3526 = 2360.
My actual points total was 1960. Doesn't seem to add up unless my maths is wrong or I've misunderstood something.
|
On March 08 2011 08:29 SDream wrote:
Edited: In the diamond examples the difference before match is only 5 points taking in consideration the high possibility of different tiers.
I'll relook the bronze-silver one and look here and edit again.
If it helps my Bronze division was absolute garbage. Maybe there are several tiers within Bronze. According to ScRanks my division was ranked around 2500th in Europe among Bronze leagues with a terrible win percentage and low average points.
|
United States12224 Posts
On March 08 2011 07:40 SDream wrote:Show nested quote +On March 07 2011 23:37 Candles wrote:Hmmm Strange that no one has done this yet. Anyway, here we go - 1816 (assuming this is points I had before the match) (1840 after) - 1271 (Before the match) (1259 After) - Broodling November - Bronze - SIlver - Furinax Mu - 2173 (before) (2161 after) - 793 Hope that helps. If not let me know what to change  Oh BTW Im on the European server. Just realised that will affect your calculations. (before) 3087 - 2966 = 121 Again,pretty close to 63*X (126)
This has a Bronze player playing against a Silver player though, so we have to assume there's some league offset at work.
Player A had (1816 + 1271) - 2959 = 128 adjusted points in Bronze.
So Player B had (2173 [points] + 793 [unspent bonus]) - 2959 [max bonus pool at the time] = 7 adjusted points in Silver.
That would mean that a 128 adjusted points Bronze player is being matched against a 7 adjusted points Silver player. Difference of 121 points. Could be that both players are in the same division tier respective to their leagues, with only the league modifier separating them? If that's 150, then it's a difference of 36 points.
|
United States12224 Posts
On March 08 2011 08:29 SDream wrote:Show nested quote +On March 08 2011 08:14 Excalibur_Z wrote:
Yeah definitely, it has to calculate how many points the game will be worth before points are actually awarded, which means comparing MMRs from the beginning.
That gap is pretty huge, more than I would have expected. 29 point difference, sure okay. 71 point difference... I've gotta think that's a separate tier. Edited: In the diamond examples the difference before match is only 5 points taking in consideration the high possibility of different tiers. Edit7: Considering that players can "eat" their bonus points if they lose enough points without being demoted, than it is interesting to think that the math on some players will have this distortion, so how could we guarantee this has never happened? o.o
That is a great point. If you get placed, then win 20 (10+10) points and lose 20 points, then win 20 (10+10) points again, you've actually spent 20 bonus pool and not 10. However, I think this fixes itself when players change leagues. For the Bronze-Silver matchup that I posted my thoughts on, the Silver guy had only 7 adjusted points, so maybe it was entirely possible that at some point he went into the negative range. That's okay though because as long as we have the total maximum bonus pool, we can determine accurately what the adjusted points are at any time based on how much unspent bonus pool they have left over.
|
I am an example of a player that can't be trusted:
Right now I have -21 adjusted points, but I had -70(a little more, maybe 73) back some months when I was already checking it. So my adjusted points don't tell my real "adjusted points", cause the "spent bonus pool" of the formula can't be determined anymore. It's probably higher than -21 though, As my losing streak of 7 games could be 70 points eatten, maybe even +50 o.O
Anyway... this method is good, I am just saying to watch out and not trust the data as 100% accurate =/
Ideally we'd have smurf accounts and check it ourselves carefully. Latin America key is only 15$ for 2 months, I am thinking on buying one just to test this stuff ^^'
|
I don't see what the problem is there SD. Points are different from MMR specifically in the fact that MMR will go negative but points do not. While your points should be visibly lower if they were meant to better reflect your MMR I don't see why the system would care when it doles out points based on your visible points at the time.
|
United States12224 Posts
On March 08 2011 09:21 SDream wrote:
I am an example of a player that can't be trusted:
Right now I have -21 adjusted points, but I had -70(a little more, maybe 73) back some months when I was already checking it. So my adjusted points don't tell my real "adjusted points", cause the "spent bonus pool" of the formula can't be determined anymore. It's probably higher than -21 though, As my losing streak of 7 games could be 70 points eatten, maybe even +50 o.O
Anyway... this method is good, I am just saying to watch out and not trust the data as 100% accurate =/
Ideally we'd have smurf accounts and check it ourselves carefully. Latin America key is only 15$ for 2 months, I am thinking on buying one just to test this stuff ^^'
Oh I see what you're saying now. If you're at 0 points then how do you know what your actual adjusted rating is? I could have 0 and go -8 -10 -5, or I could have 0 and go -12, -14, -8, -4, -9, -11 and my spent bonus pool wouldn't change because I haven't won anything, yet I pretty clearly would be in the negative adjusted points range. All I see is that I'm 0. That's definitely a valid concern and why we need to get as much data as possible because we'll start seeing accuracy in volume.
For this particular collection method (only games where both player and opponent gain and lose 12 adjusted points) I think it offsets that a bit. Because the max bonus pool is so large, that gives us a huge point range to work with. Anomalies may present themselves when players are around 100 displayed points or something, but I think players around 1000-2000 are probably "safe" in terms of accurate data since their adjusted points will probably be in the positive and they're getting points based on MMR anyway. Definitely the more data that we get, the better, though.
|
On March 08 2011 02:50 Mendelfist wrote:Show nested quote +On March 05 2011 10:04 Excalibur_Z wrote:
Hey guys, in order to help us with data collection, anytime you play a 1v1 match where you won 12 points (not including bonus pool) and your opponent lost 12 points, post all the relevant data in this thread: I said it before, but for some reason 12 point wins/losses (for me) are strangely rare. I don't know why. Here are some stats taken from my gold matches at http://www.lysator.liu.se/~john/history.htmlSomething is going on here, but I don't know what. You would think that even matches would be most common, but they don't seem to. Losses: ![[image loading]](http://www.lysator.liu.se/~john/losses.png) And wins: ![[image loading]](http://www.lysator.liu.se/~john/wins.png)
Could the reason for 12 points wins/losses is so rare be that the game consider the players even before the game is played. But after seeing the result it changes its mind and says the winner was a little better and just gives him 11 points instead?
|
So I've been trying to estimate division tiers of my opponents over the last 6 games I've played. I also went back a few weeks and checked the previous games I had recorded to get the divisions of the players and about half of them had been promo/demoted.
I'm not sure if these numbers are right but for those of you looking into this I wondered if you can in some way back up these numbers of find them useful in some way.
_Division_ - _Offset_ - note
Unknown Plat div - 2 - (-48 AdjPnts at time of game) Demoted to Gold Div: Ladranix Psi
Unknown Silver div- 0 - (203 AdjPnts at time of game) Promoted to Gold Div: Turaxis Pepper
Unknown Silver div- 2 - (81 AdjPnts at time of game) Promoted to Gold Div: Kagg Echo
Gold div Nexus Mars - 2
Silver div Infestor Zed - 3
Silver div Daggoth Quest - 1
Silver div Tarsonis Charlie - 1
Gold div Queen Mars - 0
Gold div Sakai Iota - 2
Silver div Sakai Iota - 1
Gold div Adun Victor - 2
Silver div Arbiter X-Ray - 1
Silver div Corruptor ETA - 2
|
United States12224 Posts
Could you elaborate a bit on the information presented here? I see a Gold and a Silver Sakai Iota, was that just a typo? What do the numbers for offset signify? The adjusted points were yours or theirs? What was the point change for you and the opponent in these games?
|
Oops that was a typo. Those adjusted points were for them, I probably shouldn't have included those players but I thought it might somehow be interesting or useful to note that they changed leagues (after about 40 games it appears). The numbers for the offset are my estimation based on a little math and a list of assumptions. If the numbers seem to be right I will share my methodology, if not then I don't want to waste anyones time. haha
Here I'll fix the list and include a few more variables for you
My AdjPnts - (My MatchPnts : Op MatchPnt)-_Division_ - _Offset_ - Their AdjPnts
127 - (-9 : +11) - Gold div Nexus Mars - 2 - 140
118 - (+12 : -7) - Silver div Infestor Zed - 3 - 117
130 - (+11 : -10) - Silver div Daggoth Quest - 1 - 170
132 - (-10 : +13) - Silver div Tarsonis Charlie - 1 - 272
122 - (-12 : +10) - Gold div Queen Mars - 0 - 57
110 - (+14 : -13) - Gold div Sakai Iota - 2 - 100
124 - (-12 : +14) - Silver div Drone Charlie - 1 - 150
112 - (-10 : +11) - Gold div Adun Victor - 2 - 80
102 - (-11 : +12) - Silver div Arbiter X-Ray - 1 - 241
91 - (-13 : +14) - Silver div Corruptor ETA - 2 - 167
I should note that I am in silver division Zergling Beta, which I am pretty sure has no divisional offset. Game #2 on that list is probably the most interesting considering we are both silver league with essentially identical points at the time of the match yet he only lost 7, and my MMR is already fairly stabilized. He had played only 22 games at that point.
|
United States12224 Posts
On March 10 2011 07:27 DiDigital wrote:
Oops that was a typo. Those adjusted points were for them, I probably shouldn't have included those players but I thought it might somehow be interesting or useful to note that they changed leagues (after about 40 games it appears). The numbers for the offset are my estimation based on a little math and a list of assumptions. If the numbers seem to be right I will share my methodology, if not then I don't want to waste anyones time. haha
Here I'll fix the list and include a few more variables for you
My AdjPnts - (My MatchPnts : Op MatchPnt)-_Division_ - _Offset_ - Their AdjPnts
127 - (-9 : +11) - Gold div Nexus Mars - 2 - 140
118 - (+12 : -7) - Silver div Infestor Zed - 3 - 117
130 - (+11 : -10) - Silver div Daggoth Quest - 1 - 170
132 - (-10 : +13) - Silver div Tarsonis Charlie - 1 - 272
122 - (-12 : +10) - Gold div Queen Mars - 0 - 57
110 - (+14 : -13) - Gold div Sakai Iota - 2 - 100
124 - (-12 : +14) - Silver div Drone Charlie - 1 - 150
112 - (-10 : +11) - Gold div Adun Victor - 2 - 80
102 - (-11 : +12) - Silver div Arbiter X-Ray - 1 - 241
91 - (-13 : +14) - Silver div Corruptor ETA - 2 - 167
I should note that I am in silver division Zergling Beta, which I am pretty sure has no divisional offset. Game #2 on that list is probably the most interesting considering we are both silver league with essentially identical points at the time of the match yet he only lost 7, and my MMR is already fairly stabilized. He had played only 22 games at that point.
That second game is a good data point. While your MMR was close to his points, his MMR was pretty far below yours. I wonder what other level of players he's playing. The third game, your MMR is a little above 170 Silver (11 points gained), but does that mean your MMR simply jumped 60-70ish from your last game, or is the 170 guy simply in a weaker division? Either is possible. Is there some kind of typo regarding your adjusted points between games 3 and 4? You started with 130 and should have gained +11 ending at 141, but your next game shows you at 132.
Game 5, 122 Silver points vs 57 Gold and you lost 12 meaning your MMR is at that level. That's a possible league offset of 75 assuming his tier is the same as yours, but of course we need more data to confirm. Data points like this are very important.
Game 7, you had 124 points and played against a 150 Silver player and lost 12. Possibly a different tier, possibly within the variance range.
Game 8, you had 102 points and your MMR was a little below your opponent's 241 Silver points. He gained 12 so his must have been pretty close to your 102. Strong evidence of different tiers here, if not one then two tiers of separation. Variance range still naturally to be determined, but if it is +63 per tier and he's 2 tiers away, his unmodified points would have been about 115 which is close to your 102.
|
There is a missing game between games 3 and 4. It isn't in my match history but I can say I got -9 points and he got +10. He was an 87 AdjPnt gold player, and I estimated a 0 division offset.
Also in your second sentence don't you mean his MMR is pretty far BELOW mine?
In the second game I assume the player is in a high offset division. He only has 22 games played so I think the power of the division offset is much heavier in weighting in terms of the number of points he gets from the match. In this case we can either believe that he was simply a lot higher rated than me, thus my jump in mmr next game, or we can believe he is rated about equally to me but he is in a high offset division. In that case we also have to assume that the next player is in a division with one level of offset, which brings his points back down to mine (he also has to have a somewhat lower MMR than me).
The strongest evidence of tiers is the fact that of all the players I have played in my league only one had a significantly lower number of adjust points than me.
|
Why!? why!? why?!? Why have the players among me in my bronze division some played less than 50 games totally on rank with me and i've played now over 200 games to come from 70+ and now down to 30, while the top of my list have played ca. the same amount of matches as me but are top10. check the difference. argh
![[image loading]](http://img153.imageshack.us/img153/9421/wrryy.th.jpg)
Uploaded with ImageShack.us
|
In lower leagues, particularly bronze league, it is very easy to rack up points quickly by playing competition at a higher level than other bronze players. In your case you are playing players around your skill level, while the players with more points in less games are playing players with a significantly higher skill level.
|
United States12224 Posts
On March 10 2011 08:26 DiDigital wrote:
There is a missing game between games 3 and 4. It isn't in my match history but I can say I got -9 points and he got +10. He was an 87 AdjPnt gold player, and I estimated a 0 division offset.
Also in your second sentence don't you mean his MMR is pretty far BELOW mine?
In the second game I assume the player is in a high offset division. He only has 22 games played so I think the power of the division offset is much heavier in weighting in terms of the number of points he gets from the match. In this case we can either believe that he was simply a lot higher rated than me, thus my jump in mmr next game, or we can believe he is rated about equally to me but he is in a high offset division. In that case we also have to assume that the next player is in a division with one level of offset, which brings his points back down to mine (he also has to have a somewhat lower MMR than me).
The strongest evidence of tiers is the fact that of all the players I have played in my league only one had a significantly lower number of adjust points than me.
Errrrr yeah, I did mean his MMR was below yours. I'll fix it.
I don't think that's necessarily evidence of a different division tier though. If he hasn't played many games then his MMR is probably still highly volatile, so his MMR is pretty far from your points (and therefore his own points).
|
But if his MMR is far from my points then I shouldn't have earned 12 points for the match. I could have a much higher MMR, but the rest of the matches don't seem to support my MMR being significantly higher than my points. Also it is at least peculiar that the player would reach the same number of points as me in less than half as many games, despite having a significantly lower MMR and a .500 winning%.
|
United States12224 Posts
On March 10 2011 09:16 DiDigital wrote:
But if his MMR is far from my points then I shouldn't have earned 12 points for the match. I could have a much higher MMR, but the rest of the matches don't seem to support my MMR being significantly higher than my points.
Oh yeah that's right. His MMR would have to be around your points, which is why you got 12. Your MMR could be higher which is why he lost 7, but that's unlikely because the rest of your games are pretty close to that level. I wasn't processing what his MMR would be properly, don't mind me, I'm distracted.
So yes offsets would explain it, good call.
|
I'll update every game I play here:
https://spreadsheets.google.com/ccc?key=0AoPq9zdgd7owdFFUZFIyNmtlVXJxWEtmQ1ZlSEZWeEE&hl=en&authkey=CODQx50L
Don't expect too much more than 4 games a month though 
AP = adjusted points
w/l = how many you won/lost in that game (-24<->+24)
Division names are in Portuguese (on sc2ranks they'd be in English)
There's a reason I think my current division is rank A, I would bet it is right, but could be wrong A means one tier bellow S, the top tier on any league.
If I can get any conclusions then I'll make a new post, but I won't make more posts to advertise it ^^
I'll play one game now just to not make it pointless to have done it before playing any games, I just had the idea and wanted to work on it first XD
Edit: game 1 played.
It would be cool to have more players following my model if you have the time and patience to write things down.
Cya.
|
|
|
|
|
|
EPT![[image loading]](http://img153.imageshack.us/i/wrryy.jpg/)
