(Un)Official Smogon University PO Stats -- August 2011

[XienZo]

Originally Posted by Fat Antar

http://www.smogon.com/forums/showthread.php?t=85957

Tbqh, the method described there isn't necessarily the "1 out of 20" teams either.

Rather, the formula used, (1 - (0.5)^(1 / 20)), just means that Pokemon that are above the cutoff will, out of 20 random teams, have less than a 50% chance of not showing up at all.


In practice, for a pokemon exactly on the OU cutoff, a 3.41% cutoff means that if you battle exactly 20 random teams a day, and if you play for D days, you'll run into the pokemon on an average D/2 days.

However, that's not the same as the pokemon being in 1 out of 20 teams, and I'd argue that the current definition is LESS useful than the simplified 1/20 = 5%.

The main difference is that, a 5% cutoff means that if you play against T random teams, you'll run into the pokemon on an average T/20 of those teams.

That's a lot more direct and useful than the current definition, which only states that when you play against exactly 20, and only 20, teams, you have a 50% chance of running into the pokemon.

The current definition also doesn't differentiate between if you run into the pokemon 1 time or 20 times during that 20 match streak, so while it can tell you you'll run into the pokemon 50% of your exactly-20-match-long runs, it doesn't even tell you how many pokemon you'll run into on average during those.


I apologize if I didn't explain that well, but I think you know about probability well enough to get the gist of what I'm saying, and how a 5% (or 4% or 2.5% or w/e) cutoff would both be more meaningful and simpler than our current system.

[MaestroXXVI]

the june percentages add up to 600% and include missingnos, while the august percentages only add up to 511% and dont...maybe a little math can make up for that:

first, calculate the missingno-less percentage of usages from june
(3641820-8590)/3641820*600% = 598.585%

assuming august's missingno percentage is the same as june's, we can calculate the average multiplier we need to get to go from stats not including unused pokes to stats including them: 598.585/511.029 = 1.17133

Thus, we should multiply august's given percentages by 1.171 to provide us with stats more in line with the previous months', no?

EDIT: checking that quickly means that venusaur, swampert, chandelure, mienshao, and hippowdon would remain OU, while tornadus, darmanitan, cloyster, and machamp would still drop.

[Her0]

Wow, I have been looking for these stats for a while now. This is excellent!

They do seem weird though, especially in the OU section. I had no idea Gliscor and Scizor were that high. I guess Gliscor could be used so much because of all the sand sweepers. The game definitely seems weather centered with Politoed, and Ttar all in the top 8 though. Interesting, interesting.

[Byrn Dragonstone]

Just a quick note on some of the scewed stats. In general, via the method used to determine said stats, Defensive/Offensive pivots will see a rise, whereas sweepers and cleaners will see a decrease. This is due to that fact that a pivot is literally there to come in and allow another pokemon to pivot off of it, for example Scizor's U-Turn. Thus Pivots will appear in the stats more often, whereas sweepers and end game cleaners will see a drop. But this is mostly conjecture, and it assumes rage quits, or taking out an opponent without using all of your Pokemon. For instance you sweeping a team with your Landorus, and thus your Excadrill never came into play.

[Antar*]

Originally Posted by Fat MaestroXXVI

Thus, we should multiply august's given percentages by 1.171 to provide us with stats more in line with the previous months', no?

This is certainly an imperfect solution, given that

Originally Posted by Fat Byrn Dragonstone

the method used to determine said stats, Defensive/Offensive pivots will see a rise, whereas sweepers and cleaners will see a decrease.

A cursory glance at the stats shows that, indeed, leads, weather-abusers, scouts and pivots all have gotten a bit of a boost from this method of stat calculation (just one more piece of data: keep in mind that the average number of pokemon per battle for OU is 10.2). Still, it's not a *bad* option, IMO.

I really don't think there's any way to reconcile the old and new ways of calculating the stats.

Since the stats CAN'T be calculated the old way for August (and, presumably, September), we need to decide whether to (1) arbitrarily weight August vs. May/June to come up with new usage-based tiers, (2) discard August's (and the first half or more of September's) stats and instead mesh May/June with data generated from an updated server (either Innocent Criminal's patch, when he develops it, or the 1.0.30 official update) or (3) discard May/June, and do two more months of stats with this method (or a modified method that can be done using the existing raw data) before generating new tiers.

[fatsnorlax]

Charizard is above stuff like Manaphy, Salamence, and Arceus-Steel in Uber usage...

[XienZo]

Originally Posted by Fat Antar

This is certainly an imperfect solution, given that



A cursory glance at the stats shows that, indeed, leads, weather-abusers, scouts and pivots all have gotten a bit of a boost from this method of stat calculation (just one more piece of data: keep in mind that the average number of pokemon per battle for OU is 10.2). Still, it's not a *bad* option, IMO.

What if you try to correct for the increased number of leads?

For example, tally up the usages into six categories; 1 category for the usage of a pokemon first (i.e. a lead), another category for the second pokemon used by a team in battle, and so on, until you have:

u_1,u_2,u_3,u_4,u_5, and u_6, where u_x is the total number of usages of a pokemon used xth in the battle.

For example, if there were 10 teams that revealed all of their pokemon, 7 that revealed only 4 pokemon, and 3 that only revealed 1 pokemon, it'd look like:

(3+7+10),(7+10),(7+10),(7+10),(10),(10)

and the total amount of the usages would be the sum, 91

Then, wouldn't you be able to account for the differences by recording what position each pokemon was sent out in (1st through 6th) and using the appropriate weight?

For example, if out of those, T-tar appeared 10 times as a lead, and 2 times each in the rest of the position, you could use:

10/20+2/17+2/17+2/17+2/10+2/10

and then divide by 6 to get the total usage %.

to account for the fact that early-game pokemon appear more often in general (like, how leads in this case were represented twice as much as the 6th pokemon to show up, and thus usages as leads are weighted half as much).

[Avira]

Originally Posted by Fat Antar

Since the stats CAN'T be calculated the old way for August (and, presumably, September), we need to decide whether to (1) arbitrarily weight August vs. May/June to come up with new usage-based tiers, (2) discard August's (and the first half or more of September's) stats and instead mesh May/June with data generated from an updated server (either Innocent Criminal's patch, when he develops it, or the 1.0.30 official update)

I would prefer option 2, as data has been thrown out since we have gotten usage statistics for BW, which in my opinion, is the ideal way to compile the stats for a competitive community because there are many uncompetitive teams at the lower part of the ladder, and their inclusion can change the fate of many hovering around the cut-off percentage. Meru expressed this sentiment in a simpler and better way than I did earlier.

I also wouldn't mind option 1 as a temporary measure to get rid of Pokemon like Celebi, that really have no place in UU, and then waiting for option 2 to become usable.

[Antar*]

Originally Posted by Fat Malvira

I would prefer option 2, as data has been thrown out since we have gotten usage statistics for BW, which in my opinion, is the ideal way to compile the stats for a competitive community because there are many uncompetitive teams at the lower part of the ladder, and their inclusion can change the fate of many hovering around the cut-off percentage. Meru expressed this sentiment in a simpler and better way than I did earlier.

Would that opinion change if we could institute the rating cutoff (by using current ratings for August, end-of-month ratings in the future)?

[HuntSaboteur]

I think you should create NU with the current RU stats, but leave the OU and UU unchanged.

[Antar*]

Originally Posted by Fat XienZo

What if you try to correct for the increased number of leads?

[...]

This assumes all teams are equal, specifically that each team will have the same likelihood of sending out x pokemon in a battle for all x.

In reality, teams employ different strategies. Teams that rely on heavy switching are more likely to have higher numbers of pokemon appear in the battle. Teams that rely on paving the way for one of two or three pokemon to set up and sweep will average a lot less.

What you're suggesting is that "Trainer used less than six pokemon" implies that the odds of that trainer having the sixth pokemon be x is "u_6(x)/sum(u_6)," which I am not convinced is true.

[Avira]

Originally Posted by Fat Antar

Would that opinion change if we could institute the rating cutoff (by using current ratings for August, end-of-month ratings in the future)?

If that can be done, then I wouldn't mind. I don't actually have a problem with only collecting stats for Pokemon that show up during a match. I actually prefer this to what had been done by Doug and Rising Dusk, but I didn't clarify that. I just didn't like the idea of including all battles in the statistics as I believed that it would be going backwards since player ratings have been taken into account ever since we received statistics for the fifth generation.

[Texas Cloverleaf]

Once we update to 1.0.3 it won't matter anyway as all pokemon from both teams are listed on the log at the beginning of the battle.

[Antar*]

Originally Posted by Fat Malvira

If that can be done, then I wouldn't mind.

It's not clear whether it can be done or not. There are currently 652 pages (40 per page, 274 with usage over 1000) of users for OU Standard alone. Unless I'm able to get access to a raw text file of player rankings(I'm actually considering doing a little bit of Qt programming to try to trick the server into giving me one), I don't really see what I could do to get those ratings.

[Mario With Lasers]

Originally Posted by Fat Malvira

If that can be done, then I wouldn't mind. I don't actually have a problem with only collecting stats for Pokemon that show up during a match. I actually prefer this to what had been done by Doug and Rising Dusk, but I didn't clarify that. I just didn't like the idea of including all battles in the statistics as I believed that it would be going backwards since player ratings have been taken into account ever since we received statistics for the fifth generation.

They weren't, Rising Dusk would collect every single battle statistic, even if the player had -12425 rating.

Besides, I still believe we should use the player's rating at the moment of the battle (if Innocent Criminal succeeds in developing the patch), or use none at all. Frankly, there's so much shit that makes PO's rating system less than ideal that I don't think we should even bother that much with weighted stats.

Originally Posted by Fat Texas Cloverleaf

Once we update to 1.0.3 it won't matter anyway as all pokemon from both teams are listed on the log at the beginning of the battle.

I set up a server with 1.0.30 to test this and the teams didn't appear on the server-side battle log.

[Avira]

Originally Posted by Fat Mario With Lasers

They weren't, Rising Dusk would collect every single battle statistic, even if the player had -12425 rating.

I'm pretty that Rising_Dusk didn't count everything, and I just sent a PM and am waiting for confirmation of this.

[waterwarrior]

Originally Posted by Fat fatsnorlax

Charizard is above stuff like Manaphy, Salamence, and Arceus-Steel in Uber usage...

Salamence never really fit into Ubers nearly as well as Chomp, and is constantly overshadowed by pretty much every other Dragon.

Arceus-Steel and Manaphy... fucking noobs. You'd think Manaphy's usage would be so much higher with Tail Glow being godly now...

However, just to point out the other side of the argument, with Sun, Specs Charizard with Rapid Spin support can hurt things, but then you realize... you're running Charizard in Ubers.

[Antar*]

I, ladies and gentleman, am bad. ass.

I downloaded the PO source and compiled my own special version. In this super-sexy version, whenever the program gets called on to generate a row for the rankings table widget, it also prints the rankings to the console, where it is easily copy-pastable.

It'll be tedious, yes, but I can make myself a text file of the current rankings of all players.

The question of whether we should use such a list or not... I guess that's what you're debating now.

In the meantime, I'm off...

[Edit: dammit--I made the server mad]

[waterwarrior]

Originally Posted by Fat Antar

I, ladies and gentleman, am bad. ass.

Yes. Yes you are.

[Antar*]

I have the ratings. The server kept booting me for being "overactive." After about the third or fourth time, Rare Power PC'd me to tell me to quit spamming the chat.

el. oh. el.

In any case, I gotta edit out all the "Street Pokemon" stats and some redundant entries, and then I gotta modify my script quite a bit, but in a few days, I should have "Malvira-style" usage stats for y'all.

[Chieliee]

FUCK YEAH

you're amazing

[Antar*]

Innocent Criminal would laugh if he could see what I did.

[Joel]

I like the way Antar compiled th statistics for August.

If in a battle a Pokemon wasn't used, it shouldn't be counted towards usage statics.

Choosing a Pokemon to be on a team and actually using a Pokemon on your team are not the same thing.

If I beat someone with only one or two 'mons then that means those 'mons are better than most, and should be ranked as such, no matter how much it might skew other stats.

Of course you will see a rise in leads, because they lead. If that's all you need to beat another person, then that's just tough luck for the lesser Pokes out there.

Although I do believe all battles should be counted no matter what the ranking of the battler might be. The battle still happened. It could be one of those rare instances where a +1500 person gets paired to battle a -1400 person. Would you just count half the battle because one person sucks? Or would you tell the good battler his/her battle doesn't matter because their opponent was crap? Really, that would never make any sense either way, anxd there would eventually be a lot of annoyed people.

[Limi]

Really neat, and I'm interested how new UU (when it comes ito play) will turn out. Finally Swampert will find a place for itself in UU!

[PK Gaming]

Antar is this you IRL?