/cdn.vox-cdn.com/uploads/chorus_image/image/64749068/1139719074.jpg.0.jpg)
It’s July 17th.
That means we’re in the dog days of Summer and bloggers such as myself start to get a bit bored drumming up the same stories of the offseason over and over again. So rather than do that, I decided to go totally off-book.
I was on Twitter yesterday and saw this tweet from Corey Sznajder (All-3-Zone tracker extraordinaire) in which each Carolina defender was assigned a Pokemon. That sounded like a fun exercise and so I decided to give it a go myself. But rather than subjectively assign a pokemon that felt like it described a player well, I decided to take this totally pure and childlike endeavor and absolutely ruin it by making it as numerical and objective as possible. And that’s why I decided to make Pokemon Similarity Scores.
A “similarity score” is a pretty commonterm that doesn’t really mean the same thing depending on where you’ve seen it. Here’s an example of Bill James’s (creator of Moneyball) similarity score. But all of the methods have one thing in common, which is that you need to compare similar stats. And that’s where I started taking some pretty incredible intellectual liberties.
I took this dataset from Kaggle in order to get some Pokemon-based data. They had 6 categories that seemed reasonable enough to match — HP (health), Attack, Defense, Special Attack, Special Defense, and Speed. I went through about a dozen possible interpretations of each one of these stats, but then I remembered that it’s July and I chose this topic because it was supposed to take LESS effort. So here’s what I decided for each category:
HP = time one ice per game (TOI/GP): HP is the amount of damage you can take, much like the endurance it takes to play heavy minutes
Attack/Defense = even-strength offense/defense (EVO_AA & EVD_AA): Attack is equivalent to offense, and defense is equivalent to ... well ... defense. They use the same word for that one. And the reason I’m only using even-strength is because “Special Teams” should be included in ...
Special Attack/Defense = powerplay offense/shorthanded defense value (PPO_AA & SHD_AA): Okay, I’ll admit, this one is a cop out. My other idea was that this would by “low-danger” because special attack is long-distance. I also considered using hits to split hem up because regular attack/defense are “physical.” But this is a definite win semantically.
Speed = faceoff/transition value and penalty drawing value (GAR_Zones and GAR_PD): Faceoffs and transitions help you get the first shot at attacking — anyone who’s played Pokemon knows that the higher Speed Pokemon attacks first — and penalty differential is one of the highest correlated metrics to actual “speed.”
Almost everything here was from evolving-hockey except for the speed metrics which are from Corsica. After calculating the z-scores for the pokemon dataset and the hockey dataset, I found the sum absolute difference from every one of the 151 pokemon across all categories for each skater. The then I divided that number from the highest number in the dataset (~23 SDs Brandon Pirri vs. Chansey) and subtracted it from 1. The result is the percent of the total potential deviation that was avoided. Ex: Kevin Rooney and Wartortle have an 88% Similarity score — there was only 12% of the potential deviation between their z-score profiles.
A couple of notes on the results: 1) This is using only the 151 original Gen 1 pokemon, 2) several pokemon have a LOT of matches. These are two things that I may fix over the course of the next few days, and if/when I do, the results of this article will become outdated. I also may not do that. But, my deadlines are Wednesdays and this is a hockey blog in July so you get this half-baked, but serviceable, result here. Oh, also Chansey is ridiculous and kinda ruins the HP variable and I may just remove it.
So, enough messing around with jargon. It’s time for what you guys all came for. What are the Devils results?!?! Well I’ll give you guys a few of my favorites.
Blake Coleman: Pidgeot
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18315828/pidgeot.png)
This was actually the highest match in the entire roster and it feels great to me. Two years ago, Blake Coleman was a Pidgey, just an annoying little pigeon. Last season, he became a Pidgeotto after getting a little offensive punch like maybe a Wing Attack, but also still being really annoying by repelling potential rushes with whirlwind-esque forechecking. And this year he finally graduated to a genuinely intimidating offensive force with excellent hair — Pidgeot. Nico and Palmieri also graded in as Pidgeots which is even more of an endorsement of Coleman’s excellence this past season.
Coleman’s full similarity report included Golbat and Persian as runner-ups.
Jesper Bratt: Beedrill
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18315840/beedrill.jpg)
Bratt’s Beedrill match followed only Coleman-Pidgeot in Similarity Score. This one also seems apt — Beedrill is a winged insect with stingers on either side which makes him perfect for the dual-winger, Bratt. Beedrill looks super intimidating and if it’s his move watch out because he’s good on the attack. However, if you dive deep, you see Beedrill is a pretty easy KO and Bratt’s even-strength defense is likely the worst on the team since entering the league.
Bratt’s full similarity report included Nidorino and Dragonair as runner-ups.
Travis Zajac: Clefable
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18315861/clefable.jpg)
Zajac’s Clefable is middle of the pack in Similarity score, but it’s my next most favorite one so it goes here. Clefable has been around since the first generation, but is still in the overused tier of Smogon. With high health, defense, and a moveset that includes Soft-Boiled, Clefable is a really tough KO that can be used effectively to counter some elite attackers. Similarly, Zajac is a reliable player that is still serviceable in a lot of areas. He’s a minute muncher that slows things down against Crosby’s and the like so that studs can come out well-rested after. Clefable’s is only slightly below average in attack and speed and above in everything else. This super-versatile, high-usage, oldie is a natural Zajac.
Zajac’s full similarity report included Dewgong and Lickitung as runner-ups.
Other Notables:
Taylor Hall: Victreebel — elite attacker — so-so defensively.
Other notable Victreebels: John Tavares, Steve Stamkos, Auston Matthews, Brayden Point
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18315884/Dash__5_.png)
P.K. Subban: Poliwhirl — he’ll give you the opportunities with speed and endurance, but it remains to be seen if the wrath will return.
Other notable Poliwhirls: Shayne Gosthisbehere, Kevin Shattenkirk
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18315886/Dash__6_.png)
Andy Greene: Clefairy — same idea as Zajac/Clefable except he’s older so he’s even worse.
Other notable Clefairies: Deryk Engelland, Roman Polak, Cody Ceci, Jay Bouwmeester
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/18315887/Dash__7_.png)
Some other big names I haven't mentioned include ...
Severson: Lickitung
Butcher: Parasect
Simmonds: Cubone ...
and if you want more you can just check out the Tableau I made for this very purpose! It’s also embedded below.
Thanks for reading and feel free to share any interesting ones that you find in the comments below! Feel free to offer constructive criticism for how to improve the method, but I’ll likely ignore it seeing as this is a very unimportant endeavor. Enjoy!