|
metadata

Guild Rumble #33 has begun. Good luck, everyone!
|
|
|
metadata
> *Originally posted by **[gfore25r](/forums/727068/topics/1750479?page=1#12872653)**:*
> > *Originally posted by **[gfore25r](/forums/727068/topics/1731819?page=1#12836967)**:*
> > I know it was mentioned before, but it looks like margin of victory is taken into account in [ranking movements for rumble](https://sites.google.com/view/siegebot/rumblebot). As some lines cross for a loss, particularly looking at comparable ranks following a win-loss-win pattern. I can only think of margin of victory being the cause. It could be argued that it is too small of an effect, and that it should result in a greater swing than it does presently, but nonetheless it seems to play a minor role in the rankings and only for losses.
> >
> > Update:
> > So it looks like it does take into account win margin. Both Suite Deliverance and SS Sphinx won their previous matches, and then lost match 3. Suite Deliverance lost by 12.5k while SS Sphinx lost by 7k. Suite Deliverance went from 2nd to 14th, while SS Sphinx went from 3rd to 12th. Suite Deliverance went down 5020 in MMR and SS Sphinx went down 4764. Since the caliber of the opponent was much stronger for Suite Deliverance, it must be the margin that is driving this difference. The net difference in MMR is 256, and net difference in win margins was 5.5k, results in each point of MMR being equivalent to 20 helmets in rumble.
>
> Another datapoint, both New World Order and Life After Death lost their 2nd match after winning their first. New World Order lost by 7.2k while Life After Death lost by 7.7k. New World Order went from 13 to 37, while Life After Death went from 12 to 38. New World Order went down 4287 in MMR and Life After Death went down 4385. The net difference in MMR of 98, and net difference in win margins of 558, results in each each point of MMR being equivalent to 5.7 helmets in rumble. This appears to be different than the previous results, which may indicate a tiered threshhold with varying rates, suggesting that closer matches have greater affect on MMR movement. Or it could suggest a change in the formula used to calculate changes to MMR from the last time a comparison was made.
Another crossing datapoint, both Pickle Rick and Scruffy's Closet lost their 3rd match after coming off a win in their 2nd match and losing their first. Pickle Rick lost their 3rd by 9.4k while Scruffy's Closet lost by 0.9k. Pickle Rick went from 14 to 43, while Scruffy's Closet went from 16 to 42. Pickle Rick went down 4404 in MMR and Scruffy's Closet went down 4003. The net difference in MMR of 401, and net difference in win margins of 8.5k, results in each each point of MMR being equivalent to 21 helmets in rumble. This appears to confirm the first set of results, which had wider disparity in the win margins Perhaps if you rule out a backend change, there is a tier difference when losses are greater than 8k, and those that are less get a lower rate, and thereby each helmet has a greater impact on MMR.
*Update:*
The final equation for determining the change in MMR on a loss with a Delta less than 1000 is 210 x (ln MMR Rank) - Delta/4 - 4248, otherwise 210 x (ln MMR Rank) - Delta/24 - 4248. The net MMR can change by 4100 at rank 2 to 2100 at rank 20000. For a win, it is simply 543 x MMR Rank^0.242 which is 600 points at rank 2, and 6000 points at rank 20000.
|
|
|
metadata
Thank you for your vigilant research and useful information. Is there any indication that their opponents rank have any impact on victory or MMR points?
|
|
|
metadata
> *Originally posted by **[crizz1231](/forums/727068/topics/1763014?page=1#12903123)**:*
> Thank you for your vigilant research and useful information. Is there any indication that their opponents rank have any impact on victory or MMR points?
Data seems to support that conclusion, but the two issues are intertwined.
**Nail-biter:**
MMR Rank | Change | Final Score Delta
-|-|-
5|-3967|142
31|-3588|510
36|-3596|696
51|-3543|841
**Close:**
MMR Rank | Change | Final Score Delta
-|-|-
2|-4337|4016
15|-3825|3062
19|-3896|5171
28|-3685|2222
75|-3591|3456
**Blowout:**
MMR Rank | Change | Final Score Delta
-|-|-
10|-4449|11009
16|-4148|10110
17|-4071|8552
21|-4230|12469
26|-4322|15534
33|-4157|12438
42|-4110|12609
55|-3827|9349
65|-3734|9609
73|-4041|15345
Going further, and using my previously discovered MMR modifier of 1/21 and 1/5 of the Final Score Delta, and subtracting that back out, one can back calculate the equation for the base MMR for each rank. Therefore, the final equation for determining the change in MMR on a loss with a Delta less than 1000 is 210 x (ln MMR Rank) - Delta/4 - 4248, otherwise 210 x (ln MMR Rank) - Delta/24 - 4248. The net MMR can change by 4100 at rank 2 to 2100 at rank 20000. For a win, it is simply 543 x MMR Rank^0.242 which is 600 points at rank 2, and 6000 points at rank 20000. Hope this helps!
|
|
|
metadata
> _Originally posted by **[crizz1231](/forums/726779/topics/1763014?page=1#12903123):**_
> Thank you for your vigilant research and useful information. Is there any indication that their opponents rank have any impact on victory or MMR points?
|
|
|
metadata
I’d like to see dual BGE/single BGE/dual BGE/ single BGE as a pattern again. I was hoping for a sole disguised rumble.
|
|
|
metadata
> *Originally posted by **[gfore25r](/forums/727068/topics/1763014?page=1#12903280)**:*
>
> Going further, and using my previously discovered MMR modifier of 1/21 and 1/5 of the Final Score Delta, and subtracting that back out, one can back calculate the equation for the base MMR for each rank. Therefore, the final equation for determining the change in MMR on a loss with a Delta less than 1000 is 210 x (ln MMR Rank) - Delta/4 - 4248, otherwise 210 x (ln MMR Rank) - Delta/24 - 4248. The net MMR can change by 4100 at rank 2 to 2100 at rank 20000. For a win, it is simply 543 x MMR Rank^0.242 which is 600 points at rank 2, and 6000 points at rank 20000. Hope this helps!
Sorry but I'm a bit confused by that passage.
Any chance for an Excel calculation instead?
Thanks.
|
|
|
metadata
LOL that is excel. I'll make it a little more forum friendly with my edit below:
> *Originally posted by **[werenarc](/forums/727068/topics/1763014?page=1#12905854)**:*
> > *Originally posted by **[gfore25r](/forums/727068/topics/1763014?page=1#12903280)**:*
> >
> > change in MMR on a loss:
> > Final Score Delta less than 1000 = 210 x Natural Logorithm (MMR Rank) - Final Score Delta/4 - 4248
> >
> > -Or-
> > Final Score Delta greater than 1000 = 210 x Natural Logorithm(MMR Rank) - Final Score Delta/24 - 4248.
> >
> >
> > change in MMR for a win = 543 x MMR Rank^0.242
>
> Sorry but I'm a bit confused by that passage.
>
> Any chance for an Excel calculation instead?
>
> Thanks.
>
|
|
|
metadata
> *Originally posted by **[gfore25r](/forums/727068/topics/1763014?page=1#12906992)**:*
> LOL that is excel. I'll make it a little more forum friendly with my edit below:
>
> change in MMR for a win = 543 x MMR Rank^0.242
So we won at MMR rank #49 and according to that formula, we should have gotten 1392 MMR points?
But we actually got 1482 instead?
|
|
|
metadata
Yeah, I think there may be something in there for streaks or historical strength of opponents. Anyway, here is another highly correlated excel equation for the spaghetti model:
change in MMR for a win=1000xEXP(0.0092xMMR Rank)
For your example, 1570. So average them, lol? Bam 1481!
|