Data Visualising metagame trends (v2)

Approved by celticpride

Traditionally, usage data has been analysed in month-by-month slices, as and when new statistics are published. This is great for looking at short term changes, for example the impact of a particular centralising Pokemon being banned from a tier. What about longer term trends? You need a visual tool. I have written v2 of the tool I wrote a year ago. It can be found at the URL below:

http://pokestat.org.uk


I haven't played properly since DPP, so I'm very curious to see what people who have more context of the various metagames in the modern game can come up with using this tool! Please use and abuse it and post your findings back in this thread! If you want some ideas, some of my exploration is in the spoiler section below:

Take the most used Pokemon last month, Landorus-Therian. I noticed how its usage spiked a lot in April - why? Perhaps one of it's major counters has been removed - so I stuck Ferrothorn on the graph and noted how it's usage has dropped this month, but again, why? To a regular, the answer is obvious, Volcanion. It's the secondary effect that I find most interesting, whereby the introduction of Volcanion indirectly caused Landorus usage to spike!



Consider the question: Did the introduction of the primal forms of Groudon and Kyogre have an impact on the Ubers metagame? Select Gen 6 Ubers and plot Groudon, Kyogre, Groudon-Primal and Kyogre-Primal on the graph. When the Gen 3 remakes were released, the graphs remained pretty much continuous, as if everyone pretty much just replaced the originals with the Megas. So perhaps there was little impact, or perhaps you can throw a few more Pokemon onto the graph and prove me wrong?



Again, I'm not familiar with Gen 6, so please go ahead explore those ideas and your own. Let us know what you find out! I wish I could post this in a more general thread, as it covers more than just Gen 6 OU, but this seems the most appropriate place that I have permission to post in.

Happy stat hunting!
(Remember - correlation doesn't imply causation, but speculation is always fun!)

The new version addresses a number of problems, and pieces of feedback:

  • I now cover the full set of generations and tiers covered in the usage stats
  • No data gaps (the previous version interpolated as it was missing data from certain months)
  • New published stats will be live on this site within an hour of being published
  • More extensible (the backend is storing both "real", "raw" stats and also stats cut at particular ladder rankings, which could be exposed relatively easily)
Technical details (read: it's so 2010):
  • Github repository
  • The backend is a rails server:
    • I am familiar with the technology so it was easy for me to put this together in a couple of days
    • The data model and use case suit the server / SQL backend very well (I spend most of my work time dealing with timeseries data kept in SQL databases, which sped things up)
    • cron configured to run a loader (which runs in the rails environment, again making things easier) once every 30 minutes, and only loads newly published data
  • The app is hosted on Digital Ocean, their pre-canned rails with postgres means I can go from a development environment to a hosted service in about an hour
  • The frontend uses ChartJS to plot, as well as some simple jQuery and bootstrap
Feel free to raise any bugs in this thread, or in the issues section of the Github repository! You might find that this site doesn't work well on mobiles, or screens with low resolution.
 

Martin

gamer boy
is a Forum Moderatoris a Live Chat Contributoris a Contributor to Smogon
Moderator
Out of interest I plugged the two big Ground-types (Lando-T and Garchomp) into there to see how the meta favored them, and the data is actually pretty jarring.
image.jpg

Landorus-T
Garchomp

IDRC about you seeing my tabs but try and focus more on the data lol
The basic pattern is exactly what I expected to see; Garchomp is used less than Lando-T throughout XY, gets overtaken by Garchomp in early-mid ORAS and regains its position as dominant Ground-type in November's stats. Great. However, the sheer difference in usage during XY is utterly astounding, and you can see that up until around May people were pretty much exclusively using one or the other as opposed to both, as evidenced by how Lando usage falls as Chomp rises. What does surprise me is that Lando usage begins rising right as Garchomp overtakes it while Garchomp keeps rising. This is indicative of one of two things:
  1. A rise in the utility of defensive Ground-types
  2. The general rise in popularity of "typespam" cores other than BirdSpam occuring around that time (i.e. Bisharp+Weavile, Thundurus+Manectric and Lando+Chomp offense)
My suspicion is that it is more indicative of the former, as iirc GroundSpam's moderate popularity came about more recently than May of last year, and also because if my memory serves me well Dark and Elec spam were used more around september-december of last year as opposed to May (which is further supported by the fall in Lando-T as Garchomp rises due to the latter faring better against the increasingly-popular Bisharp). What I am most interested in, however, is the shift in November. I want to say that it is because of the rise in sand, but due to the general unreliability of Excadrill and Tyranitar's usage stats as a representation of that (people still insist on using Choice Scarf Excadrill as their spinner for some reason) it is difficult to confirm this based on usage.
 
Okay so I'm aware this isn't something people are really unaware of but hoopa-u's usage and tyranitar's usage were very closely linked together (obviously because hoopa necessitates pursuit trapping and ttar is premier pursuit trapper), but even after the ban, ttar's usage has continued to climb. I believe that the continued rise of pursuit trapping is due to volcanion being so common due to it being everyone's shiny new toy. My hypothesis (using fancy words ik) is that in a few months we'll see a similar trend with volcanion and ttar as volcanion appreciates pursuit trapping. (Again I know this isn't some profound idea but it's cool to see proof)
 
Approved by celticpride

Traditionally, usage data has been analysed in month-by-month slices, as and when new statistics are published. This is great for looking at short term changes, for example the impact of a particular centralising Pokemon being banned from a tier. What about longer term trends? You need a visual tool. I have written v2 of the tool I wrote a year ago. It can be found at the URL below:

http://pokestat.org.uk


I haven't played properly since DPP, so I'm very curious to see what people who have more context of the various metagames in the modern game can come up with using this tool! Please use and abuse it and post your findings back in this thread! If you want some ideas, some of my exploration is in the spoiler section below:

Take the most used Pokemon last month, Landorus-Therian. I noticed how its usage spiked a lot in April - why? Perhaps one of it's major counters has been removed - so I stuck Ferrothorn on the graph and noted how it's usage has dropped this month, but again, why? To a regular, the answer is obvious, Volcanion. It's the secondary effect that I find most interesting, whereby the introduction of Volcanion indirectly caused Landorus usage to spike!



Consider the question: Did the introduction of the primal forms of Groudon and Kyogre have an impact on the Ubers metagame? Select Gen 6 Ubers and plot Groudon, Kyogre, Groudon-Primal and Kyogre-Primal on the graph. When the Gen 3 remakes were released, the graphs remained pretty much continuous, as if everyone pretty much just replaced the originals with the Megas. So perhaps there was little impact, or perhaps you can throw a few more Pokemon onto the graph and prove me wrong?



Again, I'm not familiar with Gen 6, so please go ahead explore those ideas and your own. Let us know what you find out! I wish I could post this in a more general thread, as it covers more than just Gen 6 OU, but this seems the most appropriate place that I have permission to post in.

Happy stat hunting!
(Remember - correlation doesn't imply causation, but speculation is always fun!)

The new version addresses a number of problems, and pieces of feedback:

  • I now cover the full set of generations and tiers covered in the usage stats
  • No data gaps (the previous version interpolated as it was missing data from certain months)
  • New published stats will be live on this site within an hour of being published
  • More extensible (the backend is storing both "real", "raw" stats and also stats cut at particular ladder rankings, which could be exposed relatively easily)
Technical details (read: it's so 2010):
  • Github repository
  • The backend is a rails server:
    • I am familiar with the technology so it was easy for me to put this together in a couple of days
    • The data model and use case suit the server / SQL backend very well (I spend most of my work time dealing with timeseries data kept in SQL databases, which sped things up)
    • cron configured to run a loader (which runs in the rails environment, again making things easier) once every 30 minutes, and only loads newly published data
  • The app is hosted on Digital Ocean, their pre-canned rails with postgres means I can go from a development environment to a hosted service in about an hour
  • The frontend uses ChartJS to plot, as well as some simple jQuery and bootstrap
Feel free to raise any bugs in this thread, or in the issues section of the Github repository! You might find that this site doesn't work well on mobiles, or screens with low resolution.
This is great! Would it be possible to do the same with moveset data or to have different graphs for different ELO brackets?
 
Awesome tool!

It should be noted that some data in here is before the mega split, so megas show 0% usage before that point in time. The normal form must be used to get a mostly accurate picture of the megas usage prior to the split.
This is so fucking cool
Thanks guys :) The mega split change is due to the way that stats were collated at the time? I actually find it pretty cool to see it change into the compositions. The majority that I've found the mega has almost entirely replaced the original, but seeing the more gentle splits like TTar andTTar-Mega is nice!

Lando T + Garchomp
Nice analysis of the XY / ORAS portion. The answer to the Novermber question is easier, plot Garchomp-Mega onto the graph and you'll find the effect is much reduced if you imagine summing Garchomp + Garchomp-Mega. What made Garchomp so popular coming into ORAS? (This is all a little astounding to me, I lived through the DP Garchomp ban!)

Okay so I'm aware this isn't something people are really unaware of but hoopa-u's usage and tyranitar's usage were very closely linked together (obviously because hoopa necessitates pursuit trapping and ttar is premier pursuit trapper), but even after the ban, ttar's usage has continued to climb. I believe that the continued rise of pursuit trapping is due to volcanion being so common due to it being everyone's shiny new toy. My hypothesis (using fancy words ik) is that in a few months we'll see a similar trend with volcanion and ttar as volcanion appreciates pursuit trapping. (Again I know this isn't some profound idea but it's cool to see proof)
Nice demonstration! I wonder if there's a certain amount of people experimenting with other TTar sets going on there as well, now that they perhaps don't need Pursuit locking up a 4th move as much? You probably know better than me! Let's check it out in a couple of months time and see if you were right!

This is great! Would it be possible to do the same with moveset data or to have different graphs for different ELO brackets?
ELO brackets will be super-simple. I'll see if I can knock that together this weekend! :D

As for moveset data, there's quite a lot! I'm curious as to what you might want to be comparing here. It'll be difficult for me to find time to implement something very general like "Plot the number of Lando-T with Impish:252HP, vs the number of Rotom-W with Choice Scarf" - but if you want something more specific, please do elaborate, so that I can see what I can do :)
 

Martin

gamer boy
is a Forum Moderatoris a Live Chat Contributoris a Contributor to Smogon
Moderator
As for moveset data, there's quite a lot! I'm curious as to what you might want to be comparing here. It'll be difficult for me to find time to implement something very general like "Plot the number of Lando-T with Impish:252HP, vs the number of Rotom-W with Choice Scarf" - but if you want something more specific, please do elaborate, so that I can see what I can do :)
Something like Sand Rush Exca v.s. Mold Breaker Exca would be a pretty good way to do it for situations like sand, and I think that abilities and items will be easier to plot by than EV spreads due to amount of variation in the latter due to things like custom spreads and speed creep.
 
As for moveset data, there's quite a lot! I'm curious as to what you might want to be comparing here. It'll be difficult for me to find time to implement something very general like "Plot the number of Lando-T with Impish:252HP, vs the number of Rotom-W with Choice Scarf" - but if you want something more specific, please do elaborate, so that I can see what I can do :)
I wasn't thinking about comparing, but simply moveset data of mons throughout time. Items and abilities should be easy too, though spreads would be trickier.

And you've implemented ELO already, that's so awesome!
 

Karxrida

Corruption of Shadows
is a Community Contributor Alumnus
Nice analysis of the XY / ORAS portion. The answer to the Novermber question is easier, plot Garchomp-Mega onto the graph and you'll find the effect is much reduced if you imagine summing Garchomp + Garchomp-Mega. What made Garchomp so popular coming into ORAS? (This is all a little astounding to me, I lived through the DP Garchomp ban!)
Maybe TankChomp as a set? Mega Metagross' steady decline started around the time Garchomp got a huge spike usage (which was around the time of Mega Metagross' suspect).

I can't remember when TankChomp became a thing and haven't yet compared other Pokemon stats, but the correlation is still interesting.
 
So I wanted to get a visual representation of just how much a stall meta has stagnated. I've used two images to demonstrate, the first is the five man Sable core and the second are some mons that are pure stall mons not in this core. As you can note, the clear trend of this first five set is clear that they're used very often in tandem, having only one month where Chansey/Skarm/Sable don't follow similar lines. Amoongus and Quagsire for their parts don't follw any particular trend until 09/2015, which I can plainly guess is the start of this five man core. As you can see, past this instance they are almost a unit, with the amount of outside stall teams in 1825+ probably being represented by the 3% gap between chansey to sable.

The honest truth is it's a bit more than the difference between Chansey and Sable. Sable does have some balance teams but chansey shouldn't be found elsewhere. Well, the more common chansey is on other styles, the less stalls outside sable you'll see, as suggested by this data. As it is in the last data point, there is less than a percent difference, meaning less than 10% of stall in 1825 have a mon other than Sable. At this error, we can probably assume that it isn't stall, but chansey being on a different style.


This second image compares sableye to Slowbro, Venusaur, and Celebi. I wanted to use a mon like Jirachi or heatran but their ability to be used on multiple styles so easily skews data. Also tangrowth is so new I'm not sure I could actually find realistic and decent data. I guess you could say this chart shows exactly what you'd expect: Celebi, which did drop to UU, Venusaur, Slowbro all trend basically opposite to sableye. Slowbro does have that stint before the five man squad was adapted where he trends similar to sableye, but notice how the huge spike in sable's popularity chopped about 5% of Slowbro's usage.




These charts represent what any stall builder already knew, Sable is literally choking out the stall meta. The answer is pretty easy: Sable is more oppressive to stall teams than he actually is to offense teams, in the fact that stall rarely has the power to destroy sable. So quite realistically, Stall either runs sable or loses to sable. It is kinda (very) unhealthy to stall's general health, but there doesn't seem to be much to do about it.

So yeah, this was mostly to show that the numbers can now prove, beyond a reasonable doubt, that sableye-m is capable of choking out an entire subset of a style (aka non-sable stall). Take what you want from this, I'm just praying Gen7 shafts this P.O.S. mon.
 
I dont exactly think its fair to say that sableye is "choking out" other styles of stall as much as i would say that without sableye, stall reallly isnt very viable at all. Stall teams just get worn down far too easily with hazards up to really work. The decline in stuff like venusaur is also due to other metagame trends screwing it over like sand for example. Celebi is on the decline because people are realizing how utterly awful it is compared to stuff like amoonguss or tangrowth. Stall has really reached an optimized state where it manages to not autolose to much, and be very effective against an average ladder player. Other stall megas like venu and alt just tend to produce far less consistent and effective builds, and thus have lost quite a bit of usage. Thats just my opinion on the matter though
 
I dont exactly think its fair to say that sableye is "choking out" other styles of stall as much as i would say that without sableye, stall reallly isnt very viable at all. Stall teams just get worn down far too easily with hazards up to really work. The decline in stuff like venusaur is also due to other metagame trends screwing it over like sand for example. Celebi is on the decline because people are realizing how utterly awful it is compared to stuff like amoonguss or tangrowth. Stall has really reached an optimized state where it manages to not autolose to much, and be very effective against an average ladder player. Other stall megas like venu and alt just tend to produce far less consistent and effective builds, and thus have lost quite a bit of usage. Thats just my opinion on the matter though
It's fair on the grounds that Stall existed fine before sable. It might not have been strong, but it certainly was viable. I don't get this "stall tends to get worn down by hazards" argument. Stall in metas without sable deploys hazards and offense deploys hazards. Stall hazard deployers normally have more longevity, as do their clearers. Hazards have existed since gen2? and Stealth rocks since gen4, and stall has been viable in multiple metas since.

The venusaur data in particular coincides too perfectly for me to say "his decline is to sand" and not sable. No, his general usage is because of sand. Compare most of Venu to Sable and mega venu to mega sable lines. They're near opposite. I agree Celebi wasn't the greatest choice but things like Heatran, Jirachi, Clefable and other 'stall mons' have too much usage elsewhere to show good trends.

I don't believe this state of stall is optimized at all. In fact, I feel the team is actually very weak to easy threats, mainly physical fires like Stall breaker talon and ZardX. Taunters with any ability to 2hko sable wreck this team and Stallbreaker Heatran is a perpetual destroyer to this team like no other. Fairies also run it down pretty hard. Sable stall inherently leaves quite an interesting pattern of threats unchecked so as to keep sable safe.

The thing is, what you say about hazards is VERY TRUE in context of a stall team vs Sableye. Stall teams inherently have two patterns: Bulky mons and lots of utility. Meaning they lack offensive presence and rely on utility moves to make up for that. We could categorize utility moves into "Self serving" and "Aggressive" utility (example, Wish, heal bell into self serving, toxic into aggressive). Sableye is able to block (almost) all aggressive utility and his team is able to use his own. Most notably, stealth rocks and spikes. And stall has no way to easily clear out a sableye. Really, stall doesn't clear out a lot of threats and sable is able to prey on that fact. Recovery, bouncing back stall's main form of damage/general pressure, and good utility moves himself means that he's putting his own team at a huge advantage in any stall vs stall fight.

If it were possible to do an experiment, I would design a defensive builder to face off against an offense builder under the understanding that each will build similar styles. Restrictions being you can't just pile on stallbreakers and shit (make it a meta offense team of some sort). My bet is the winrate between a Sable stall and a Non-Sable Stall in that situation is rather similar, if not favoring non-sable stall (they'll be more concerned about wholly walling the meta rather than supporting sableye).

If you run that same experiment with Sable stall v non sable stall, Sable stall's win rate would be close or above 90%. Because while Stall has no real issues with hazards, having only one team face that means in the eventual 2-300 turn fight that can ensue, the non-Sable stall team has taken probably 4-500% HP hazard damage and the sable maybe 50%. I think it comes down to this: Sable stall is REALLY good vs other stall teams. If Stall isn't meta, than Sableye would be dropped from the team for a more wholesome defensive mega. This is my belief on why Stall has drifted to him so heavily.
 

Infernal

Banned deucer.
I was toying around with this tool and ended up wanting to share some random thoughts based on any known trends I could come up with. Used 1825 ELO for these:
There was talk here on just how common cores with things like Lando-T, Latios, and Jira currently are. These guys are some of your typical glues and the utility/coverage over threats they offer make them really appealing choices. I plugged in their names into this tool to see what would pop up and this graph was the result. Lando-T + Latios + Jira follow each other upwards quite closely during the past three months. This makes sense seeing how popular they are together right now. U-turn + strong breakers is a common strategy now, and Lando-T + Jira are some of the best Pokemon to use with this style. Add a Latios as your catchall check to things like Keld and you have a core capable of covering many big threats (ranging from fairies like Mega Diancie to physical threats like Exca). I wouldn't be surprised to see these three continue to rise with each other in the coming months.
Like someone already said, as Hoopa-U became more popular, things like Weav and Tar did as well. Even after its ban, those two still remain very popular. Sharp isn't in the same situation however. It's declining and has been for a while. This makes sense to me because Sharp doesn't perform as consistently compared to those other dark types, performing the trapper role less reliably and lacking the utility things like Tar provide. Pokemon like Keldeo and Terrak are also popular, however I do think the strong competition Sharp faces from other dark types could be the main factor. There's something else I wondered: dark is such a powerful type, could choice items have any influence on their usage? Tar is a known choice user. Hoopa-U was a known choice user, with Specs being the one people generally considered as 'too much' to handle. Even Weav is starting to see more use with CB, and these items all enhance how threatening these guys are. Sharp can't really use choice items to the same effect as them, so could this (along with other factors) have anything to do with its decline?
The drop in offensive electrics is something people have discussed many times before. The graph mainly shows them as declining over the months. This makes sense with sand's prominence and the popularity of Mega Latias/defensive grass types. Threats like Manaphy have become less common too, so maybe that's a factor? Offensive electrics currently aren't very hyped as a whole. One interesting thing according to the graph is the spike in Thund-I's usage near the end. I do think Thund-I is still strong despite issues like SR and the popularity of threats like Weavile. Like bludz said here, teams relying on cores with a mix between Lando-T, Rotom-W, Keld, Jira, and Latios can often end up weak to Thund-I, so we'll see whether the increasing popularity of these builds does anything to influence its usage. Added Hippo to this graph mainly because I think its decline may also correlate with the fall in electrics as shown. One of Hippo's big advantages over things like Lando-T is dealing with offensive electrics better. With electrics declining, it's not unreasonable to say there is less appeal to using Hippo as recent months have shown.
Grasses are another thing people have discussed over the months so that's something else I looked for. Tangrowth and Amoonguss have both been popular during recent months and the graph illustrates this, showing how both have risen. Mega Venu however is experiencing a decline as shown. You can also see Loom's usage fall as those defensive grass types rise. The same reasons for Mega Venu's decline have been discussed many times before (sand interfering with Synthesis, burns reducing its effectiveness, and so on). Tangrowth and Amoonguss have Regen to make some of these issues less annoying and also cover many of the same threats as Mega Venu without using up your mega. Along with other traits (like Spore and so on), all of this has made them generally more attractive options and I can see them continuing to rise.
Zard-X is something I haven't seen around much for a while. We all know the rise of Lando-T is a factor, but I decided to see how Zard-X correlated with Chomp's usage. This is because Zard-X generally threatened teams much more during the time where Tank Chomp was the premier bulky ground. With the graph, you can see Zard-X follows Chomp quite closely: as Chomp's usage declines, so does Zard-X's. Is this mainly because of Chomp being overtaken by Lando-T as the preferred bulky ground? Although there are several other factors at play here, that's still something I found interesting to think about.
Wanted to see how the Lati's graphed up and the results were predictable. Latias consistently declines while Latios stays strong. Mega Latias also increases in usage mostly. Latios' power and Mega Latias' higher bulk seem to eclipse most of what Latias can bring to the table. I like how closely Latios' usage follows Tar's. This makes sense, as the more Latios is used the more you'd expect Tar to be around.
 
Last edited:

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

Top