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

Status
Not open for further replies.
cdbk said:
So I had a decent amount of free time, so I thought:


Tier Shift predictions for November and December!

These predictions have been made using 2 key factors:
  • Viability in the tier (and consequently metagame shifts)
  • The usage stats from October

OU

Rises:
hoopa-unbound.png
gliscor.png
latios-mega.png
(♂)
rotom-wash.png
gyarados-mega.png


Drops:
gengar.png
garchomp.png
(+
garchomp-mega.png
please low ladder)
mew.png
tapu-fini.png
skarmory.png
stakataka.png
dragonite.png
mimikyu.png
manaphy.png
charizard-mega-y.png
marowak-alola.png
ninetales-alola.png
lycanroc-dusk.png


UU

Rises:
seismitoad.png
moltres.png
quagsire.png
kommo-o.png
(probably drop to RU once the January Stats are revealed)

Drops:
celebi.png
entei.png
arcanine.png
forretress.png


RU

Rises:
necrozma.png
yanmega.png
sigilyph.png


Drops aka I beg you, please stop clutching onto these:
596MS.png
719MS.png
banette-mega.png
142MS.png
181-m.png
323-mega.png
764MS.png
452MS.png
768.png
009.png
781MS.png
232MS.png
dugtrio.png
695MS.png
135MS.png


NU

Rises:
335.png
type-null.png
481.png
altaria.png
haunter.png
miltank.png
passimian.png
silvally-steel.png
silvally-water.png
435.png


Drops:
573.png
693.png
golbat.png
344.png
237.png
028-a.png


Now let's just hope that this will be accurate looking at you RU
Click to expand...
Stop being so hopeful
 
Lordemort said:
How is Mega Ampharos good?
Click to expand...
Punish standard defensive cores that consist of mega steelix/milotic/snorlax or spdef mon by itself since thanks to STAB Electric attacks and focus blast it destroys mega steelix and milotic singledhandlely, and can just volt switch out of the spdef mons that can outlast it, and momentum is always nice for a team.

It can be pretty tough to cover for balance/bulky offense due to its raw bulk and power. 4 attacks is the set that makes the best out of this mon in the meta tbh.

This isnt the place to answer metagame questions tho, for RU questions go here or if you have more questions regarding mega ampharos or other RU Pokemon.
 
This is getting ridiculous, can we at least get some kind of update or timeline on when we might be getting november stats? from what i understand they were supposed to be available at the beginning of january because we've had a whole month of waiting and we're already a week in with no update at all. Myself and probably the entire lower tier community is just sick of being left in the dark waiting for stats to come, especially when they're incredibly important due to both SPL and the Exhibition happening right now. It's not even about wanting the stats to come sooner, if we could just get some kind of accurate and reliable timeline for the events that are happening right now as well as future stat drops, i'm sure that would appease the majority of people in lower tier communities.
 
Last edited:
^^^^Yes. Myself and others feel confused as to how five new pokemon seem to have completely dismantled the usage statistics. Is there a greater problem? As much as we all appreciate everything that goes into this, it may be time to question whether or not more responsability needs to be spread among other members in this matter. Just my thoughts and the thoughts of my friends. Not trying to belittle the amount of work that goes into this.
 
Samsuro said:
^^^^Yes. Myself and others feel confused as to how five new pokemon seem to have completely dismantled the usage statistics. Is there a greater problem? As much as we all appreciate everything that goes into this, it may be time to question whether or not more responsability needs to be spread among other members in this matter. Just my thoughts and the thoughts of my friends. Not trying to belittle the amount of work that goes into this.
Click to expand...

I agree. But unfortunately nothing is going to change until the higher-ups hear enough complaints from users
 
Antar said:
~Snip
Click to expand...

Not sure how relevant this is, and hopefully it doesn’t add to your headache.

The December stats are showing a really low number of battles across the board for all the metagames I’ve looked at. Thought it would be worth bringing to your attention given the pressure on you for tier changes etc.
 
Antar will the next "full" tier shift happen in March as originally scheduled or will it be moved to April due to the delay?
 
Those "Total battles" numbers are definitely wrong, considering the "Avg. weight/team" numbers are very much above the logical maximum of 1.0
http://www.smogon.com/stats/2017-12/gen7ou-0.txt said:
Total battles: 1159
Avg. weight/team: 1830.915
+ ---- + ------------------ + --------- + ------ + ------- + ------ + ------- +
| Rank | Pokemon | Usage % | Raw | % | Real | % |
+ ---- + ------------------ + --------- + ------ + ------- + ------ + ------- +
| 1 | Landorus-Therian | 30.12812% | 1278656 | 30.128% | 265 | 22.927% |
Click to expand...
Yeah, something in the script is still broken.
 
Drk Pwnr said:
Those "Total battles" numbers are definitely wrong, considering the "Avg. weight/team" numbers are very much above the logical maximum of 1.0

Yeah, something in the script is still broken.
Click to expand...
The Chaotician said:
Not sure how relevant this is, and hopefully it doesn’t add to your headache.

The December stats are showing a really low number of battles across the board for all the metagames I’ve looked at. Thought it would be worth bringing to your attention given the pressure on you for tier changes etc.
Click to expand...

Well...

Antar said:
There was a bug in the November run that messed up the team matchups and "Real" stats, so I'm rerunning those before posting them.
Click to expand...
 
Drk Pwnr Quote -- investigating now. The thing that borked November should have been fixed for the December run, and anyway, it shouldn't have affected the battle-count.

Update: whatever it is, it appears to have only affected singles formats, but it does appear that it affected _all_ singles formats.

Update 2: so the good news is that, like the bug that's causing me to rerun November, the usage counts themselves are unaffected. Which means that the only thing that will affect the tiers is the math that combines suspect and non-suspect ladder stats, which means that the only tier that could possibly be affected will be PU, since NU is the only tier that had suspect tests in December.

Update 3: to be clear, I know there were other suspect tests in November, but those all seem to have the correct battle counts (meaning either PS changed some log format in December (unlikely) or my fix for the November bug broke something (extremely likely)
 
Last edited:
I used the numbers in old stats files to reverse engineer some equations and figure out what the bug is.

TotalBattles * Raw% * 2 = Raw

Plugging the first few December numbers into that equation, you get
x * .30128 * 2 = 1278656 (Landorus-Therian)
y * .17859 * 2 = 757957 (Tapu Koko)
z * .15206 * 2 = 645361 (Greninja-Ash)

Taking the average of those x y and y values gives 2122052.85859. On a hunch, I divided that by the incorrect Total Battles number 1159 and got... 1830.93430422, the incorrect Avg weight/team number.

Looks like an operation was accidentally reversed somewhere.

(So, if anyone else obsessively needs to get the total battles numbers before the stats script is re-run, you can calculate it yourself like this)
 
Drk Pwnr -- it's not a math problem. It's due to this code block being called before this one. That error message was all over the log, but (1) I didn't realize it was hitting basically 100% of battles and (2) I didn't realize the execution was happening in that order.

So basically,
  • My code is sloppy and not very resilient, nor is it particularly easy to tell when something has gone horribly wrong
  • There's something somewhere else that's causing this not-very-resilient-but-has-been-working-just-fine-for-months code to break, most likely in this commit or this one.
 
Okay, I think I have it fixed. This is why unit testing is important--when I cleaned up my code in the last commit, I assumed I'd correctly translated the structure for the lines I was looking for. I didn't. If I'd had a unit test that tried this code on a sample log,it would have been immediately obvious that I'd made a mistake.

Let this be a lesson for you all in how not to do software engineering.
 
Status
Not open for further replies.
Back
Top