10 ms·

Xkcd-Style Plots in Matplotlib (2012)

325 points by TsukiZombina 8y ago

pfd1986 8y ago
Needs 2013 tag. Still fun tho.
- lrem 8y ago
  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.
emgee_1 8y ago
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.
- altairiumblue 8y ago
  https://cran.r-project.org/web/packages/xkcd/vignettes/xkcd-intro.pdf https://cran.r-project.org/web/packages/xkcd/vignettes/xkcd-...
  - emgee_1 8y ago
    Thank you very much! I will look into this.
- nestorD 8y ago
  There is a xkcd package to fulfill this need : https://stackoverflow.com/questions/12675147/how-can-we-make-xkcd-style-graphs https://stackoverflow.com/questions/12675147/how-can-we-make...
- leemailll 8y ago
  I'm curious was it easy to share macro back then?
altairiumblue 8y ago
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.
- jjcc 8y ago
  Yes. I follow him for a while. Great guy. Seems working for google now? https://www.youtube.com/watch?v=inN8seMm7UI&t=4s https://www.youtube.com/watch?v=inN8seMm7UI&t=4s
- simonh 8y ago
  Why a shame? Is there something better it’s crowding out?
  - Tarq0n 8y ago
    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.
    - billfruit 8y ago
      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.
    - galenlynch 8y ago
      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.
      - 8y ago
wodenokoto 8y ago
Is there a bootstrap or css theme to go along, so my PoC web-app can match the style of my proof of concept graphs?
- turbinerneiter 8y ago
  There is https://roughjs.com/ https://roughjs.com/ - but the style is different and it's for drawing SVG.
- nohuck13 8y ago
  We've had pretty good results with PaperCSS [1]. For bonus points: xkcd-font from the Jupyter team [2] [1] https://www.getpapercss.com/ https://www.getpapercss.com/ [2] https://github.com/ipython/xkcd-font https://github.com/ipython/xkcd-font
- meredydd 8y ago
  I put together an "XKCD" theme for Anvil that works quite nicely: https://anvil.works/blog/xkcd-style-apps https://anvil.works/blog/xkcd-style-apps
loser777 8y ago
it would be cool if Randall stopped drawing plots by hand and just started using this ;)
- ianbooker 8y ago
  I would not be surprised if he does exactly that already ;)
sharpercoder 8y ago
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.
- ianbooker 8y ago
  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? ;)
  - ericsoderstrom 8y ago
    (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.
    - coldtea 8y ago
      >Lines that make up characters in a font don't have any semantic meaning related to the words. Err, actually they do. If not the line itself, the font's overall design does. A font can look casual, another official, fun, etc. In fact this very post uses the "Humor Sans" font for the graphs for exactly this very reason. To make labels appear more casual.
  - stared 8y ago
    Not necessarily. For Comics Sans it is hard to know if it is an intentional or just sloppiness. For xkcd plots, it is clear that someone put an extra effort in making it look sketchy.
  - abhinavsharma 8y ago
    Balsamiq is software that’s an example where picking deliberately poor fonts is a common thing to do in low-fidelity prototyping. It helps because it focuses attention on higher level elements over nitpicking on details https://balsamiq.com/?gclid=EAIaIQobChMI5cTZhKzm4AIVtx-tBh3OMgtNEAAYASAAEgK6zvD_BwE https://balsamiq.com/?gclid=EAIaIQobChMI5cTZhKzm4AIVtx-tBh3O...
stared 8y ago
(2012), infuential but archaic. Now it is built-in matplotlib: https://matplotlib.org/api/_as_gen/matplotlib.pyplot.xkcd.html https://matplotlib.org/api/_as_gen/matplotlib.pyplot.xkcd.ht...
vntok 8y ago
yUML.me is awesome for hand-drawn DB schemas or to show a website's graph of interlinked pages: https://yuml.me/diagram/scruffy/class/draw https://yuml.me/diagram/scruffy/class/draw
jen729w 8y ago
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.
masukomi 8y ago
it should be noted that the code to do this is actually part of matplotlib now. The author comments about it here http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matplotlib/ http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matpl... and refers to the code in the post here as "my ugly little hack."
__blockcipher__ 8y ago
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.
- EForEndeavour 8y ago
  You're not the only one. Here are slides from a talk that Joel Grus gave at JupyterCon 2018 titled "I hate Jupyter Notebooks": https://docs.google.com/presentation/d/1n2RlMdmv1p25Xy5thJUhkKGvjtV-dkAIsUXP-AL4ffI https://docs.google.com/presentation/d/1n2RlMdmv1p25Xy5thJUh... Video of talk: https://www.youtube.com/watch?v=7jiPeIFXb6U https://www.youtube.com/watch?v=7jiPeIFXb6U
  - __blockcipher__ 8y ago
    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
panda88888 8y ago
Follow up in 2013. XKCD officially merged in Matplotlib. http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matplotlib/ http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matpl...

325 points by TsukiZombina 8y ago

pfd1986 8y ago

Needs 2013 tag. Still fun tho.

lrem 8y ago
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.

emgee_1 8y ago

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.

altairiumblue 8y ago
https://cran.r-project.org/web/packages/xkcd/vignettes/xkcd-intro.pdf https://cran.r-project.org/web/packages/xkcd/vignettes/xkcd-...
- emgee_1 8y ago
  Thank you very much! I will look into this.
nestorD 8y ago
There is a xkcd package to fulfill this need : https://stackoverflow.com/questions/12675147/how-can-we-make-xkcd-style-graphs https://stackoverflow.com/questions/12675147/how-can-we-make...
leemailll 8y ago
I'm curious was it easy to share macro back then?

altairiumblue 8y ago

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.

jjcc 8y ago
Yes. I follow him for a while. Great guy. Seems working for google now? https://www.youtube.com/watch?v=inN8seMm7UI&t=4s https://www.youtube.com/watch?v=inN8seMm7UI&t=4s
simonh 8y ago
Why a shame? Is there something better it’s crowding out?
- Tarq0n 8y ago
  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.
  - billfruit 8y ago
    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.
  - galenlynch 8y ago
    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.
    - 8y ago

wodenokoto 8y ago

Is there a bootstrap or css theme to go along, so my PoC web-app can match the style of my proof of concept graphs?

turbinerneiter 8y ago
There is https://roughjs.com/ https://roughjs.com/ - but the style is different and it's for drawing SVG.
nohuck13 8y ago
We've had pretty good results with PaperCSS [1]. For bonus points: xkcd-font from the Jupyter team [2] [1] https://www.getpapercss.com/ https://www.getpapercss.com/ [2] https://github.com/ipython/xkcd-font https://github.com/ipython/xkcd-font
meredydd 8y ago
I put together an "XKCD" theme for Anvil that works quite nicely: https://anvil.works/blog/xkcd-style-apps https://anvil.works/blog/xkcd-style-apps

loser777 8y ago

it would be cool if Randall stopped drawing plots by hand and just started using this ;)

ianbooker 8y ago
I would not be surprised if he does exactly that already ;)

sharpercoder 8y ago

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.

ianbooker 8y ago
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? ;)
- ericsoderstrom 8y ago
  (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.
  - coldtea 8y ago
    >Lines that make up characters in a font don't have any semantic meaning related to the words. Err, actually they do. If not the line itself, the font's overall design does. A font can look casual, another official, fun, etc. In fact this very post uses the "Humor Sans" font for the graphs for exactly this very reason. To make labels appear more casual.
- stared 8y ago
  Not necessarily. For Comics Sans it is hard to know if it is an intentional or just sloppiness. For xkcd plots, it is clear that someone put an extra effort in making it look sketchy.
- abhinavsharma 8y ago
  Balsamiq is software that’s an example where picking deliberately poor fonts is a common thing to do in low-fidelity prototyping. It helps because it focuses attention on higher level elements over nitpicking on details https://balsamiq.com/?gclid=EAIaIQobChMI5cTZhKzm4AIVtx-tBh3OMgtNEAAYASAAEgK6zvD_BwE https://balsamiq.com/?gclid=EAIaIQobChMI5cTZhKzm4AIVtx-tBh3O...

stared 8y ago

(2012), infuential but archaic. Now it is built-in matplotlib: https://matplotlib.org/api/_as_gen/matplotlib.pyplot.xkcd.html https://matplotlib.org/api/_as_gen/matplotlib.pyplot.xkcd.ht...

vntok 8y ago

yUML.me is awesome for hand-drawn DB schemas or to show a website's graph of interlinked pages: https://yuml.me/diagram/scruffy/class/draw https://yuml.me/diagram/scruffy/class/draw

jen729w 8y ago

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.

masukomi 8y ago

it should be noted that the code to do this is actually part of matplotlib now. The author comments about it here http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matplotlib/ http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matpl... and refers to the code in the post here as "my ugly little hack."

__blockcipher__ 8y ago

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.

EForEndeavour 8y ago
You're not the only one. Here are slides from a talk that Joel Grus gave at JupyterCon 2018 titled "I hate Jupyter Notebooks": https://docs.google.com/presentation/d/1n2RlMdmv1p25Xy5thJUhkKGvjtV-dkAIsUXP-AL4ffI https://docs.google.com/presentation/d/1n2RlMdmv1p25Xy5thJUh... Video of talk: https://www.youtube.com/watch?v=7jiPeIFXb6U https://www.youtube.com/watch?v=7jiPeIFXb6U
- __blockcipher__ 8y ago
  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

panda88888 8y ago

Follow up in 2013. XKCD officially merged in Matplotlib. http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matplotlib/ http://jakevdp.github.io/blog/2013/07/10/XKCD-plots-in-matpl...