We owe a lot to the filmmakers of skateboarding. Those whom live behind the scenes, creating the brand & image of their employer while getting little
recognition, low
pay, and poor treatment. If only there was a way
to pay these vanguards the respect they deserve. Something like an article in a niche, low-viewed online
magazine to formally recognize their unsung talents, thankless hours, unending sacrifice...
Fret no more! We're here to deliver exactly that. Well, kind of. Using a dataset I scraped from
skatevideosite.com, I've ammassed a collection of
skateboarding filmmakers, alongside the skaters they've included in their video. To be clear, I am defining 'filmmaker' as the show runner of the project: the chief creative force who is often the editor (and sometimes the actual camera operator), but not always. Additionally, skatevideosite data is user-curated, so it isn't always perfect or consistent. Read more about data quality in the Data
& Methodolgy section below.
By virtue of our data, the basis of comparison among the filmmakers will be by proxy of who they film and include
in their videos, not by their direct talents behind the camera. While this isn't the homage article they deserve, it
does provide a unique avenue of comparison among the filmmakers themselves, and illuminates some of the little communities and neighborhoods within popular skateboading.
Below each filmmaker is represented as a network graph. Each node (circle) represents a skater, sized by how many videos they were featured in by the corresponding filmmaker. The edges connect skaters that appeared in films together. Hover over each skater to see who they share connections with, and click-and-drag them around for a better view.
From the graphs above, it's clear that many filmmakers have a group of skaters that they often work with.
Sometimes this is a result of the filmers preferring to film their own crew of friends. Other times,
it's because the filmmakers work on behalf of a company, so they often film the same people.
In any case, I've always wondered which of skateboarding's filmmakers feature the most consistent cast of
skaters in their videos?
To answer the question, below I've ranked the filmers by a straightforward metric of consistency,
calculated by dividing the total number of parts a filmmaker has filmed by the unique number of skaters
they've filmed. (You can think of it as the average number of parts a filmmaker films for the skaters they film). A value of 1 for the metric, the lowest value it can take on, indicates that no skaters
had more than one part with the filmmaker. Higher values indicate that the filmer tends to film each skater
multiple times:
that they prefer to film the same group of skaters.
The results are shown in the table below:
With a max value of 3.27, no filmmaker has an overwhelming preference of skaters they desire to film.
The top 3 filmmakers, Kevin Barnett, Beagle, and Aaron Meza, head the video content for three of the most
familial teams in skateboarding (Toy Machine, Baker, and Girl, respectively). Jackson Casey, with his usual
crew from Arizona and the Midwest, barely edges French Fred out for a place in the top four. All the way at the
bottom of our table, we see Chris Ray, Bill Weiss, and Jon Holland, filmmakers for skateboarding media companies Transworld and Digital. As these companies employ a unique cast of skaters in each video, their values of 1 are expected.
Values in the middle of the table are more interesting. Dan Wolfe, who sits roughly in the middle of the table,
was the staff filmer for both Element and Real, although he filmed a number of influential independent films as well. Given the massive number of skaters he's filmed, it's somewhat surprisng seeing him sit higher in the rankings than people like Greg Hunt of Pontus Alv, people I'd just expect to be more consistent, I suppose.
Which filmers are the most similar, with respect to whom they film?
This is obviously not the first thing we think of when comparing filmers - we'd normally look at things like
the style of their filming, the cameras they use, the intricacies of their edits, etc.
However, it's an application of the data we have on hand, and provides an additional means of comparison
among filmmakers.
To answer our question, we'll borrow some embedding and dimensionality-reduction techniques from the field of information
retrieval, and apply them in a unique way to our use case (learn more about what I did in the data & methodoly section). The results are below.
Basically, each filmmaker is represented as a single dot in the scatter plot below.
The distance between each filmmaker is meaningful:
the closer two dots (filmmakers) are to each other, the more similar are the skaters that they've filmed.
Filmmakers that are further away from each other include more different sets of skaters in their videos.
Whereas our first chart gave us a view into the skater communities within each filmmaker, this chart provides us with a view into the neighborhoods amongst our filmmakers themselves.
This structure of data allows us to view clusters of similar filmmakers.
The most obvious cluster is all of the Crailtap dudes in the bottom right corner. It's not surprising to see
these guys grouped amongst each other: Crailtap skaters tend to act as a pretty tight crew, and their teams don't really change much over time. Their grouped distance from other videographers also suggests that the skaters are fairly insular and keep to themselves, preferring not to 'cross-pollinate' and appear in videos without each other too often.
About as far away as you can get from this group, nested at the very upper-left corner of the plot, Blackbox affiliates Jamie Thomas and Mike Gilbert are grouped very close together. Theres is another group of fairly insular skaters.
Towards the bottom-left of the chart, we see filmmakers with quite a bit of overlap in whom they choose to film.
If you're bored, go through the filmmakers and view their network graphs above, finding the source of intersections for yourself.
While this article isn't the homage that skateboarding filmmakers deserve, I hope it's an interesting piece that shines more light into the important work they do and some of the legacy they've left on skateboarding.
The data was collected from (the now basically
defunct) skatevideosite.com.
The data comprises videos up until July 2019. As the data in skatevideosite.com is
user-entered, If can't guarantee total accuracy. For this reason, some of our mapping from filmmakers to skaters
may not be perfect (corrections were made where identified). Additionally, not all filmmakers featued in the
dataset were included - lesser-known filmers were excluded, as were filmers with less than four unique videos
in dataset. For this reason, some important filmmakers, such as Danny Minnick, Matt Eversole, or Stacey Peralta, were left out (we're bummed, too).
To create the embeddings for each filmmaker in the scatter-plot, I used a technique called term-frequency inverse-document frequency (TF-IDF). The technique is usually used to compare documents to each other, by similarity of their diction. Basically, documents that share more words are more similar, and words that appear in fewer documents are weighed more heavily.
For our case we extend it to what we'll call SF-IFF: skater-frequency inverse-filmmaker frequency. Instead of representing documents by their words, we represent filmmakers by their skaters.
We define skater-frequency (SF) as
e.g. $$ {sf_{i,f} = \frac{\text{skater}_i \quad \text{frequency with filmmaker}_j}{\text{total skaters with filmmaker}_j }} $$
and inverse filmer frequency (IFF) as
e.g. $$ {iff_{i} = log(\frac{\text{Total filmmakers}}{\text{filmmakers with skater i}}) } $$
we create our embeddings for skater i in filmmaker j as:
$$ {sf\text{-}iff_{i,f} = sf_{i, j} \times iff_i } $$
So now, we have a numerical way to represent each filmmaker, representing the skaters that they film.
Skaters who film
the same skater will be more similar to each other ("skater-frequency"). Skaters that appear in films from a large number of filmmakers (such as Andrew Reynolds) will have less weight than rare skaters (such as Jaewoo Bae) when calculating similarity ("inverse-filmmaker frequency").
The output from our SF-IFF embedding for each skater is a vector that's too high dimensional to view, so I use a common dimensionality reduction technique called PCA to bring it back into two dimensions (read about the method here). Doing so provides us with a way to represent our filmmakers in only two-dimensions, e.g.
$$ {filmmaker_{f} = [x_1, x_2]} $$
Each dimension is a principal component with interpretations we don't need to worry about - what's important is that the filmmakers are now embedded in a space where distance is meaningful, and in one we can visualize easily (in a two-dimensional scatterplot).