Data Official Smogon University Usage Statistics Discussion Thread, mk.3

Antar

is a Battle Server Administratoris a Programmeris a Super Moderatoris a Community Contributor
Official Data Miner
No, it's just how Python is rendering 0 to that level of precision.

Stats for December are updated, but I don't think it's worth it to rerun the tier update with only 11 days left in the month, especially since only NU/PU will be affected. Quick-changes resume next month, and I'm hoping that the January run is smooth so I can get that update out quickly.
 
Right now, Gyarados's ability usage statistics in OU look like this.
+----------------------------------------+
| Abilities |
| Mold Breaker 100.000% |
+----------------------------------------+
I can already tell from Gyrados's Bulbapedia article that it always has mold breaker after it evolves, so this information is probably not very useful. On the other hand, I would like to know what percentage of Gyaradoses had intimidate before they mega evolved. Can moveset stats files show these statistics from now on?
 

Counting Sheep

formerly Volcanoheadproduction(yt)
IDK if this is the place to ask, couldnt see a simply questions and answers thread, and the rules apply from previous thread thing takes me back like 5 threads, but not to the original with the rules ?_?

I know Iv seen it somewhere before, maybe you dont do it anymore, but just incase, where can I find the stats for a Pokemons most common switch ins ?
 

Antar

is a Battle Server Administratoris a Programmeris a Super Moderatoris a Community Contributor
Official Data Miner
This might be a dumb question, but why is leads data (http://www.smogon.com/stats/2017-12/leads/) never calculated for gen 7 doubles OU? As far as I can tell, that format has no leads data files ever, despite such data files existing for gen 7 doubles Ubers and gen 7 doubles UU. Is this a bug or is there an explanation for it?
My stats system pulls data from two parts of the battle logs: the team metadata, which describes each Pokemon's set (IVs, EVs, moves, items...) and the turn-by-turn battle log itself. Leads (and checks/counters and "real" usage) are pulled from the battle log itself. Due to the range of possibilities being so much broader for doubles than singles, I never got around to writing a good battle-log parser for doubles. Consequently, no checks/counters, "real" usage or "leads." If you look at the code it literally says, "if doubles format, skip parsing the battle-log"

Edit: Stats for the month are now up.
 
Last edited:
Noted. I don't know if you plan on deleting the bug-induced files so no one falls for untrustworthy data, but if you were, here's a non-comprehensive list of which formats were incorrectly included in which months: https://pastebin.com/xgGfy4Nr (non-comprehensive because it's just VGC and Doubles Ubers/OU/UU for then-current generations).
 
Last edited:
Hey All,

I have extracted, transformed, and loaded the raw OU JSON data into SQL Server as relational data for reporting and analysis. Before I start investing (time and financially) into this personal project, I have 2 questions about some numbers:

1) What do the numbers in the 'Checks and Counters' mean? How are they calculated?

2) What do the numbers in 'Viability Ceiling' mean? How are they calculated?

Attached are screenshots of the numbers.

Thanks!
 

Attachments

Antar

is a Battle Server Administratoris a Programmeris a Super Moderatoris a Community Contributor
Official Data Miner
opw_Blade:
  1. the checks-counters numbers are:
    1. # of encounters between Swellow and Porygon-Z where one or the other Pokemon was KO'd or switched out
    2. Fraction of such encounters where Swellow was the one who was KO'd or switched out
    3. Std dev for that fraction (which can be calculated from the above two numbers)
  2. the viability ceiling numbers are:
    1. number of teams using that Pokemon (IIRC this is just the raw usage count)
    2. highest GXE of any player using that Pokemon
    3. 99th percentile GXE
    4. 80th percentile GXE
And on another note, stats for the month are now up.
 
When listing item usage, would it be possible to combine the stats for the 50% Berries (Figy Berry, Wiki Berry, Mago Berry, Aguav Berry, Iapapa Berry) into a single figure? They are functionally identical to each other but, particularly in VGC thanks to the item clause, the usage on each of them is often quite split. Having to pull out a calculator to determine their actual usage compared to other items is a pain, and even that may not always be an accurate method if a certain variety of 50% Berry doesn't show up on the list.
 
Can I bump this?
Right now, Gyarados's ability usage statistics in OU look like this.
+----------------------------------------+
| Abilities |
| Mold Breaker 100.000% |
+----------------------------------------+
I can already tell from Gyrados's Bulbapedia article that it always has mold breaker after it evolves, so this information is probably not very useful. On the other hand, I would like to know what percentage of Gyaradoses had intimidate before they mega evolved. Can moveset stats files show these statistics from now on?
Aside from this, I have two questions. Why are the 3-month stats weighted 20:3:1 and not some other ratio like 9:4:1 or 4:2:1? Second, I know that one of the reasons why stats don't weight winning teams any more than losing teams is because everyone should ideally have a 50-50 win rate. Showdown! isn't ideal, though. There are trolls who use real teams to climb the ladder and then use their troll teams once they're high enough to get weighted. So if giving weight to winning teams will stop these trolls better, then why don't we do it?
 

DoW

formally Death on Wings
is a Pre-Contributor
Can I bump this?Aside from this, I have two questions. Why are the 3-month stats weighted 20:3:1 and not some other ratio like 9:4:1 or 4:2:1? Second, I know that one of the reasons why stats don't weight winning teams any more than losing teams is because everyone should ideally have a 50-50 win rate. Showdown! isn't ideal, though. There are trolls who use real teams to climb the ladder and then use their troll teams once they're high enough to get weighted. So if giving weight to winning teams will stop these trolls better, then why don't we do it?
With regards to the trolls, I can't imagine they're common enough as to actually make a noticable difference in the stats, and certainly not a large enough difference that it's worth changing how stats are calculated for.

I've been looking at the moveset calculating code recently, and although it's probably possible to do what you want with regards to megas, it's not a simple one-liner. I'll keep it in mind when I'm looking through the rest of the codebase, though, and maybe submit a pull request for it.

Wasn't able to find the reasoning for 20-3-1, I remember reading about it but it's probably in some ancient thread in Policy Review somewhere idk
E: This seems relevant https://www.smogon.com/forums/threads/quickrises.3627841/
 
Last edited:

Users Who Are Viewing This Thread (Users: 2, Guests: 0)

Top