Doubles Ladder + Seasonal Stats

marilli

With you
is a Forum Moderator Alumnusis a Community Contributor Alumnusis a Tiering Contributor Alumnusis a Top Contributor Alumnusis a Past SPL Championis a Former Other Tournament Circuit Champion
If I read everything right you said you used the frequency of types only to create RC, then frequency of physical / special moves was seen separately and was used for your "general bulk count" that takes weighted average of physical / special defenses, then the two were multiplied? Wouldn't that overlook how certain types are almost always represented more heavily on one side of the physical / special split than the other?

Providing a Filter View option for your Google doc might let people see your results more clearly. Some kind of relevancy scale like usage (while not ideal you probably have these numbers in the spreadsheet already as opposed to using anything else), lets me look at the important numbers without sifting through Beedrill and Rattata. Probably me just being bad at Google Docs though tbh lol.
 

Level 51

the orchestra plays the prettiest themes
is a Site Content Manageris a Community Contributoris a Top Tiering Contributoris a Contributor to Smogonis a Top Smogon Media Contributoris a Team Rater Alumnusis a Forum Moderator Alumnusis a Battle Simulator Moderator Alumnusis a Past SCL Champion
If I read everything right you said you used the frequency of types only to create RC, then frequency of physical / special moves was seen separately and was used for your "general bulk count" that takes weighted average of physical / special defenses, then the two were multiplied? Wouldn't that overlook how certain types are almost always represented more heavily on one side of the physical / special split than the other?

Providing a Filter View option for your Google doc might let people see your results more clearly. Some kind of relevancy scale like usage (while not ideal you probably have these numbers in the spreadsheet already as opposed to using anything else), lets me look at the important numbers without sifting through Beedrill and Rattata. Probably me just being bad at Google Docs though tbh lol.
Hi! You're indeed correct that I have overlooked this. Along with Bughouse reminding me about X-Act's bulk metrics, in future I might look into them to adjust for type effectiveness. Consider this spreadsheet a first step, I suppose.

In terms of Filter View, I don't actually have the usage stats in the spreadsheet. The original plan was to have a page where you could input any Pokemon's name (and ability) and get its data, but unfortunately I went overboard and added every single Pokemon into there. I might re-add the calculator page, though. Realising how actually useless this thing is has knocked some of the wind out of my sails.
 

Bughouse

Like ships in the night, you're passing me by
is a Site Content Manageris a Forum Moderator Alumnusis a CAP Contributor Alumnusis a Tiering Contributor Alumnusis a Contributor Alumnus
You shouldn't be down on it at all. It's quite a good first effort. There are definitely things I'd change, and I'm happy to chat some time if you'd like. But these are long term about taking decent outputs and just making them better (such as by changing to x-act's metrics). But room for improvement doesn't mean anything is bad.

However, in the short term, at the very least please do move whatever you have going on in row 1 of the data to a separate lookup sheet. When you filter on a column at the moment, it filters the whole column, including row 1, which is what is breaking all the formulas. This is just basic usability and should not take long to fix.
 

Level 51

the orchestra plays the prettiest themes
is a Site Content Manageris a Community Contributoris a Top Tiering Contributoris a Contributor to Smogonis a Top Smogon Media Contributoris a Team Rater Alumnusis a Forum Moderator Alumnusis a Battle Simulator Moderator Alumnusis a Past SCL Champion
You shouldn't be down on it at all. It's quite a good first effort. There are definitely things I'd change, and I'm happy to chat some time if you'd like. But these are long term about taking decent outputs and just making them better (such as by changing to x-act's metrics). But room for improvement doesn't mean anything is bad.

However, in the short term, at the very least please do move whatever you have going on in row 1 of the data to a separate lookup sheet. When you filter on a column at the moment, it filters the whole column, including row 1, which is what is breaking all the formulas. This is just basic usability and should not take long to fix.
Hey, thanks! I thought I froze row 1. When I sort in Google Sheets, it seems to work for me; here's the data sorted by Relevant Bulk:


In other news, after reading up on X-Act's metrics, I realised that what I'm using is exactly the same as his statistical HP / Def / SDef; I just haven't normalised my data yet. Might do this once I have more time to. Exams coming up etc etc.
 

Bughouse

Like ships in the night, you're passing me by
is a Site Content Manageris a Forum Moderator Alumnusis a CAP Contributor Alumnusis a Tiering Contributor Alumnusis a Contributor Alumnus
Doesn't work for the view that the shared users in view-only get to see. Seriously. Just put it in a a separate lookup table. That's best practices for controlling formulas based on fixed values anyway.

Level 51 edit: didn't want to post again, but I did this
 
Last edited by a moderator:

ryo yamada2001

ryo yamada2001
is a Community Contributor Alumnusis a Tiering Contributor Alumnusis a Top Contributor Alumnus
DISCLAIMER: This is not an official ranking system currently

Hi there! It's Elise here with a fresh new ranking system. TheRealSaltyChips set up a formula and after inputting a lot of numbers for 2 days we finally came to a ranking system which takes in account all the other Doubles tours, such as DPL and SPL, as opposed to the Power Rankings which only takes in account Seasonasls.

Obviously the system is absolutely not perfect yet but we're trying our best to find better ways to improve the system and calculations. If you have any questions about the formula feel free to PM me or TheRealSaltyChips about it, as we'll be able to answer everything!

So far we've put in several tournaments and obviously this will grow over time. Hopefully the system will be refined enough so it can be used as the "main" ranking system, for suspect tests, or other crazy stuff. Right now however, we probably should just look at the numbers and see if they'll work well over time.

This is the spreadsheet which contains the rankings of 173 players. If you have any questions about the colours, and what they're used for look in the hide tag below:

In the B column, a purple name implies the player was in SPL7. A green name implies the player is on the Doubles Council.
During an active tournament, a greyed out name implies the player has been knocked out of the tour.
During any tournament, a red name implies the player has not passed Round 4 or has not played enough games to qualify for the normal multiplier.
A light blue name implies this player won the tournament, or was on the winning team for said tournament.
A rating with a different background implies this player was in the top 12 for a seasonals, or was a finalist in a tournament such as DPL or SPL.


The rankings are recorded individually on tabs as well, so you can see how well a player has developed in the tournament. Anyway if you're too lazy or do not have time to check out the spreadsheet, here's the current top 15: (last updated: 27th of July 2016)

Name - Points
1. checkmater75 - 4531
2. MajorBowman - 4136
3. Lohgock - 3938
4. Memoric - 3857
5. Stax - 3674
6. Biosci - 3464
7. Braverius - 3398
8. SamVGC - 2955
9. deoxys speed - 2749
10. Level 51 - 2690
11. Stratos - 2678
12. BLOOD TOTEM - 2646
13. Arcticblast - 2615
14. Dawgie - 2608
15. Nido-Rus - 2584
 
Last edited by a moderator:

ryo yamada2001

ryo yamada2001
is a Community Contributor Alumnusis a Tiering Contributor Alumnusis a Top Contributor Alumnus
Winning 0 SPL battlea gets you 1400+ points
?
There's a concern about that and considering the system was kinda made literally two days ago it has some flaws.
kamikaze went for a very high amount which inflated his starting value we calculated with. This made him drop not so hard as he was maybe supposed to. I think SPL experience might also be worthwhile as these players generally have experienced working with other top-tier players and are more familiar with the system and the setup which could be quite valuable to managers.

However, if you have any suggestions to improve the rankings I'd happily listen to it. Maybe we should decrease the multiplier for players who had 0 or 1 wins in the SPL as they obviously didn't do so well.
 

Platinum God n1n1

the real n1n1
is a Tiering Contributor
I'd like to see the formula you guys are using.

here is my suggestion

power ranking = points = number of total wins

done.

more complicated suggestion to account for win percentage
power ranking = points * multiplier
where multiplier equals
multiplier = total wins / total battles fought

more complicated suggestion to reward beating strong opponents and penalize losing to weak opponents
power ranking = points * weighted multiplier
where multiplier equals
weighted multiplier = [(avg points of opponents beat)*(battles won)] / [(100-avg points of opponents lost to)*(battles lost)]


with this there no need to add extra value on winning tours/advancing far as your points will increase by advancing so far as will your multiplier. Also no need to add extra value to SPL/DPL.
If you are good at math tell me what you think. If you are not good at math idgf

also I think the most complicated is the best but requires the most work and the results will not be significantly different that the first formula

edit:you'll get some small numbers so normalize on a base of 100 or 1000
 
Last edited:

Level 51

the orchestra plays the prettiest themes
is a Site Content Manageris a Community Contributoris a Top Tiering Contributoris a Contributor to Smogonis a Top Smogon Media Contributoris a Team Rater Alumnusis a Forum Moderator Alumnusis a Battle Simulator Moderator Alumnusis a Past SCL Champion
Hey Elise! I took a little time to look through the spreadsheet, and I can see where you're coming from but I just wanted to inquire about some of the specifics:
  • I'm assuming the ranking system is a variant of Elo or some other match-making system, which is for sure a cool idea I wanted to try out myself. However, what is the K-value you are using for this version of Elo? At current, the rankings don't appear to be going up or down a lot based on a win, and losses seem to be pretty harsh. I'd like to draw your attention, for example, to the case of gorex vs Mint16, who were both in the Loser's Bracket (but notably, still in the tournament) come Round 7. After their battle, Mint16 had a rating of 935 (and was knocked out of the tournament) while gorex's was 1087. Note that it is possible to be knocked out of the tournament in Round 2 with a rating as high as 906 (and a W-L of 2-4). Not sure dropping the tournament multiplier to 0.1 before hitting Round 4 is sufficient; perhaps do this gradually? Though that might outweigh the influence of the rating system in the first place.
  • The decay constants appear to cause older tournaments to have lower weightages, which is alright. However, having a high decay constant to amplify tournaments such as SPL will also amplify lower scores (hence why going 0-x gives >1400 points). SPL is weighted almost / over twice as high as other tournaments while the individual ratings don't show all that much variance due to the comparatively small number of games (and bo1s). Perhaps tweak the K-value to cause more variance, as well as lowering the weightage on SPL?
    • I've never played in SPL so I wouldn't know, but no matter how hard SPL is, it seems silly to give more points for going 0-x in SPL than for winning a Seasonal.
  • Decay is really harsh, Winter 2015 being 0.4 = ouch. Also, consider lowering the 0.1 multiplier along with the standard tournament multiplier (eg 0.08, 0.06, ...)
  • Also consider weighting decay multiplicatively instead of additively. For example, you could have Summer as 0.8, Spring as 0.8*0.8 = 0.64, Winter as 0.8*0.8*0.8 = 0.512, etc. This seems to make more sense to me since your implementation of decay is a multiplicative effect to begin with.
 

ryo yamada2001

ryo yamada2001
is a Community Contributor Alumnusis a Tiering Contributor Alumnusis a Top Contributor Alumnus
Hi guys! It's Elise here and today TheRealSaltyChips and I did some work to the spreadsheet.
- We redid the DPL starting everybody at 1000 points, which made some people drastically drop in amount of points.
- We changed the Spring 2016 decay to 0,64 (0,8 * 0,8) after Level 51's suggestions.
- We changed the Winter 2016 decay to 0,51 (0,8 * 0,8 * 0,8 rounded) after Level 51's suggestions.
- The decay of SPL7 was set to 1 so that SPL doesn't inflate ratings like crazy, and as someone above said, it's a little crazy to have someone who went 5-4 in SPL have earn more points than the Seasonals winner.
- We redid the SPL7 starting everybody at 1000.
- For DPL and SPL we added a "Playoffs" bracket, which basically awards more points for winning while in playoffs. We thought this was a logical choice due to the pressure there is during playoffs.
- kamikaze17's and Laga's decay were set at 0,3 for getting 0 wins in SPL7. deoxys speed's decay was set at 0,5 for getting only 1 win.

After all these changes the top 16 looks like following:
Name - Rank
1. checkmater75 - 4052
2. Stax - 3878 (+)
3. Lohgock - 3538
4. Memoric - 3409
5. MajorBowman - 3381 (-)
6. Braverius - 2917 (+)
7. SamVGC - 2913 (+)
8. Level 51 - 2887 (+)
9. Biosci - 2873 (-)
10. Stratos - 2795 (+)
11. Dawg - 2670 (+)
12. Nido-Rus - 2667 (+)
13. Elegyy - 2654 (+)
14. n10sit - 2405 (+)
15. Yoda2798 - 2401 (+)
16. clius - 2336 (+)

I'd like to see the formula you guys are using.

here is my suggestion

power ranking = points = number of total wins

done.

more complicated suggestion to account for win percentage
power ranking = points * multiplier
where multiplier equals
multiplier = total wins / total battles fought

more complicated suggestion to reward beating strong opponents and penalize losing to weak opponents
power ranking = points * weighted multiplier
where multiplier equals
weighted multiplier = [(avg points of opponents beat)*(battles won)] / [(100-avg points of opponents lost to)*(battles lost)]


with this there no need to add extra value on winning tours/advancing far as your points will increase by advancing so far as will your multiplier. Also no need to add extra value to SPL/DPL.
If you are good at math tell me what you think. If you are not good at math idgf

also I think the most complicated is the best but requires the most work and the results will not be significantly different that the first formula

edit:you'll get some small numbers so normalize on a base of 100 or 1000
Hi n1n1! I appreciate the criticism but a win percentage isn't as relevant as it does not take in consideration the ranking or skill of another opponent. Also I don't really see the point of creating a custom multiplier because outside of that taking a shitton of time, I have a strong feeling a set decay number is more useful and reliable. We already work with starting values of 1000.

Considering you wanted to know how the formula worked exactly; here you go!
Winner’s New Rank:
=C2+(E2/(20*(C2+E2)/2000))

Where C2 is the winner’s previous rank, E2 is the loser’s previous rank.
The winner gets 50 points added to their score when both begin with 1000, because that’s what the formula is centered around.
As the Combined score rises, the winner still receives an amount centered at 50. As the opponent’s record reaches higher than the winner’s, the winner receives more points.

Loser’s New Rank:=E2+(J2/(20*(C2+E2)/2000))

Where E2 is the loser’s previous rank, J2 is the opposite of the loser’s previous rank.
The loser standardly loses 50 points when both opponents start at 1000. With the extra (C2+E2)/2000, it balances the scoring so if both opponents have 2000, the loser still drops 50.
Just as well, as the opponent’s score reaches higher than the loser’s score, the loser will be deducted less points, and vice versa.

Bracket Multipliers:
In the first round there is no multiplier, the new ranks are pumped out based on the above formulas.
In the winner’s bracket, the winner receives a bonus to their score of 0.5% of their new score, to further separate successful players’ scores from the unsuccessful ones. Losers in the winner’s bracket lose 0.5% of their new score as well.
In the loser’s bracket, the winner’s score has no multiplier, but if they lose, they also lose 1% of their new score.
Towards the end of the tournament, winners receive 1% of their score added on, to further separate the scores of the higher players from the lower players.
This is the current system we use for seasonals, other tournaments have a different bracket multiplier to separate the more successful players from the less successful players in a smaller amount of matches.
This encourages a healthy growth to the winning players’ scores without overly inflating them.
Using this method, early losers normally score between 750-900. Middle eliminations normally have 1000-1200, and Late tournament goers can have 1300-1700. Those who go out within the first 4 rounds receive a stronger score decay than those who surpass it as well.
The bracket multipliers are different in every tournament so far because we were working on critiquing the value, it’s been everywhere from 2% for winners down to -2% for losers. We like how this newest system worked out.


Hey Elise! I took a little time to look through the spreadsheet, and I can see where you're coming from but I just wanted to inquire about some of the specifics:
  • I'm assuming the ranking system is a variant of Elo or some other match-making system, which is for sure a cool idea I wanted to try out myself. However, what is the K-value you are using for this version of Elo? At current, the rankings don't appear to be going up or down a lot based on a win, and losses seem to be pretty harsh. I'd like to draw your attention, for example, to the case of gorex vs Mint16, who were both in the Loser's Bracket (but notably, still in the tournament) come Round 7. After their battle, Mint16 had a rating of 935 (and was knocked out of the tournament) while gorex's was 1087. Note that it is possible to be knocked out of the tournament in Round 2 with a rating as high as 906 (and a W-L of 2-4). Not sure dropping the tournament multiplier to 0.1 before hitting Round 4 is sufficient; perhaps do this gradually? Though that might outweigh the influence of the rating system in the first place.
  • The decay constants appear to cause older tournaments to have lower weightages, which is alright. However, having a high decay constant to amplify tournaments such as SPL will also amplify lower scores (hence why going 0-x gives >1400 points). SPL is weighted almost / over twice as high as other tournaments while the individual ratings don't show all that much variance due to the comparatively small number of games (and bo1s). Perhaps tweak the K-value to cause more variance, as well as lowering the weightage on SPL?
    • I've never played in SPL so I wouldn't know, but no matter how hard SPL is, it seems silly to give more points for going 0-x in SPL than for winning a Seasonal.
  • Decay is really harsh, Winter 2015 being 0.4 = ouch. Also, consider lowering the 0.1 multiplier along with the standard tournament multiplier (eg 0.08, 0.06, ...)
  • Also consider weighting decay multiplicatively instead of additively. For example, you could have Summer as 0.8, Spring as 0.8*0.8 = 0.64, Winter as 0.8*0.8*0.8 = 0.512, etc. This seems to make more sense to me since your implementation of decay is a multiplicative effect to begin with.
So basically I've already talked with Level 51 about this but I'll just summarize this for people who thought the same:
Essentially it is a very very very very dumbed down version of ELO which just takes many shortcuts to do what ELO is supposed to do. It does not take in account any K-value.
I agree with the losses being too harsh while the wins can be a little unrewarding. TRSC and me applied a different multiplier on the brackets on a different seasonal and it worked much much better. However, we did this on the Winter 2015 seasonal. This would mean if we would want to redo two seasonals, which are about 700 games. I hope y'all can understand I'm not gonna input 700 games again for extremely minimal benefit.

Also marilli suggested to make it open source, so people can see the formula, but I have no idea how to do that! Sorry! If anyone could help me out on this that'd be lovely.
 

Users Who Are Viewing This Thread (Users: 1, Guests: 0)

Top