From a usability perspective, "sketchy" or "cartoony" charts actually serve a purpose. To me, they communicate "the trend is important here, not the precise numbers". This can more acurately convey communication intent as opposed to a precise rendered graph.
I think I agree on the core idea, but if you extrapolate this into font selection, for example, you would rather pick Comic Sans for more "strategic" and abstract projects? ;)
I've certainly used comic sans intentionally when presenting data. Much of this depends on audience too. A corporate board meeting is quite different from an activity with a group of rowdy 8-year-olds from a presentation in a university class from a tight-nit group of friends launching a startup.
Some comes down to power hierarchy too. Presenting up, I'm much more likely to try to appear professional. Presenting down, I go for approachable.
(Clearly you're joking) but this isn't an extrapolation. Lines that make up characters in a font don't have any semantic meaning related to the words. Lines in a graph do have semantic meaning related to the data they represent.
[edit] maybe you could make this argument for a language like Chinese. At least for characters with semantically meaningful radicals.
Funny enough, I posted an animated plot on Reddit generated using the Matplotlib xkcd style a couple of months ago and the top comments were about the choice of font - for people not familiar with the xkcd comic, this seemed to be distraction point.
Jake is a great teacher - I would definitely recommend his book to people new to Python for data science, as well as any talk by him that you can find on youtube.
Also, it's kind of a shame that matplotlib is still so deeply ingrained in the Python data ecosystem.
Assuming all you want to do is make some graphs, almost every alternative is better. Matplotlib is like the opengl of plotting libraries, there's almost no abstraction, you have to create plots in an imperative way and it's extremely verbose.
A plotting library focused on useability should let you specify what you want in a declarative manner, then get out of your way. Ggplot2 being the prime example.
It seems to be somewhat inspired by Matlab, which isn't an elegant system by any stretch. And for the people who hasn't had exposure to matlab, matplotlib seems overly convoluted.
Do you have any concrete examples for Python? As far as I know, ggplot2 is only for R. I would be interested to try out some alternatives to Matplotlib.
Am I the only one who kind of hates the "notebook" format?
Don't get me wrong, the general concept is great. Shareable, interactive code snippets are awesome. But for a demo article like this I want a single python file that I can run and immediately see the results. I don't want to have to spin up a jupyter or pylab instance (I've never used pylab so not sure if it works the same EDIT: pylab is analogous to matplotlib, not jupyter. lesson learned). I just want to run the damn code.
On a bit of a tangent here but I also hate the way Jupyter makes git diffs absolutely unusable.
Now, the above being said, the fact that sites like Github have native Jupyter functionality is awesome. It'd be a lot less painful if they'd (the author) have linked to a repo that we (I) could run.
--
For example, first I copied the initial code snippet that defines the xkcdify function. Then I ran it with python3, and realized it needed python 2 due to the urllib2 dependency (this is not a big deal since the article is from 2012). Then I ran it again with python2, and realized I don't have numpy/scipy etc installed for python2, so I pip installed those. Then I copied in the following code snippet that generates the plot. I then ran it again, and it still didn't work. Finally after a brief google search, I realized I needed to put pylab.show() at the end since I didn't have the %pylab inline or whatever that command is since I'm running it with "pure python".
Honestly, it really wasn't _that_ much effort, but I vastly would prefer to have a demo like this given as a single python file, with the dependencies clearly specified in the blog post.
Wow, thanks so much. Great talk. I felt like I was going crazy trying to get these stupid notebooks to work with their weird out-of-order execution glitches
I’ve always had a problem designing diagrams in Visio. (I work in IT so think high level network diagrams, flowcharts describing state/processes, that sort of thing.)
Last year I found the template that makes everything look hand-drawn. It changed the way my brain saw the diagram, going from “this thing must be absolutely pixel perfect” to “this is something I literally sketched on the back of a napkin”.
This allowed my brain to create the thing, without worrying about whether everything was lined up just so. Then, at the end, I can choose to flip the theme back to straight edges, and line everything up if I choose.
Actually, many times I just left it as it was. Others seemed to like the hand-drawn look, which came as a surprise.
Interesting. Is something like this available in R?
In a same vain: when producing content using latex ( in the eighties ) I had a problem that papers under construction looked way to good ( lie as if was printed) ; in order to counter the good looks of the printed paper I choose to use a typewriter like blurred ink dropped font to ensure that what was printed was a draft and still under construction; producing graphs that show intent but not precision is very useful.
In the intervening years there've also been some nice posts about this. In particular I agree with one that argued that using this style in professional setting is indeed a good idea, to emphasize that what you're presenting are hand-waved assumptions.
