Pointillism in Tableau
Despite being
nearly 4 years old, I recently discovered George Gorczynski’s blog post about
creating Pointillistic Art in Tableau. As defined by Dictionary.com, Pointillism is “a technique
of neo-impressionist painting using tiny dots of various pure colors, which
become blended in the viewer's eye. It was developed by Georges Seurat with the
aim of producing a greater degree of luminosity and brilliance of color.” For
example, here’s a painting by Georges Seurat called “Gray Weather, Grande Jatte”
Wikipedia Commons
George’s blog
includes code that reads the pixels of an image, simplifies the colors, then writes
the data to a text file, which can be easily consumed in Tableau. His work is
quite brilliant and it’s been used to create some really intriguing
visualizations, such as Curtis Harris’s visualization of Donald Trump’s tweets
(click on the image to see the full visualization).
I love how,
instead of just creating the image in Tableau, Curtis encoded each dot with
information about individual tweets. It’s both artistic and insightful, which I
think is a great combination.
So, if you
haven’t read George’s post, go check it out!
Another Technique
Shortly after
discovering George’s technique, I came across an article by Matteo Ronchetti, a data
scientist from Italy who specializes in machine learning and computer vision. The
article, titled Creating Pointillist Paintings with Python and OpenCV, describes a brilliant
algorithm he wrote in Python which converts images to pointillistic artwork. The
technique shares some similarities with George Gorczynski’s method, but has
some significant differences as well. For instance, Matteo wanted to mimic the
small brush strokes used in many pointillistic paintings. To demonstrate this,
the image below zooms in on Gray Weather, Grande Jatte.
To address
this, Matteo used small ellipses instead of dots, which adds a really nice effect
to the resulting image.
So, before I
move on, please go read Matteo’s article—it’s a brilliant and inspiring piece
of work. And, if you’d like to turn your own images into pointillistic art,
download the code from his GitHub repository and give it a go.
Preparing the Algorithm for Tableau
Any time I see
some amazing work like this, as a Tableau person, I always want to see if I can
leverage it in Tableau. This, of course, was no exception. So, after getting
permission from Matteo, I downloaded his code with the intention of exporting data
so that I could recreate it in Tableau. In the end, I had made a few modifications
such that the code would:
1) Export a CSV of Ellipse Data. The
resulting CSV has 7 columns as shown below.
Ellipse ID is a numeric identifier of each ellipse.
X and Y define the coordinates of the center of each ellipse. Length and StrokeScale define details about the length and width of each
ellipse. Angle defines the rotation
of each ellipse (measured in Degrees). And, Color defines the color, in hex, of each ellipse.
2) Write a Color Palette Text File. The file contains
the 20 colors generated by the algorithm and is pre-formatted as a Tableau color
palette so you can easily copy and paste it into your Preferences.tps file (if
you’ve never create a custom color palette for Tableau, check out the Online Help).
3) Prompt for Input File. I wanted to make
the code as easy to use as possible, without prior knowledge of Python, so I
added code that will prompt for an input file, as well as some messaging
telling the user where the output files have been written. Ultimately, I’ll
package this up so that it can be run via an executable.
The updated
code can be found on my GitHub repository.
Visualizing in Tableau
The first
thing I did was load my custom color palette. The palette file is ordered
alphabetically by color and, since my csv data has the color in it, I will be
able to sort my data by the color, then apply the palette.
Next, I’d need
to be able to plot each ellipse, rotate them, and color them. Unfortunately,
ellipses are not necessarily straightforward to create in Tableau. So I had to
use one of my favorite mathematical techniques, parametric equations (for more
about using parametric equations, see Beyond Show Me Part 3: Parametric Equations). I won’t go into a lot of detail
about how to create an ellipse with parametric equations, but I’ll just say
that the equations account for the length and width of the ellipse as well as the
rotation.
I then used a
free image found on Pixabay.
And here’s the
final visualization. Click on the image to view the interactive viz on Tableau
Public.
As I’ve done
in some of my previous posts, I think it’s really cool to see these build ellipse
by ellipse, so here’s an animation. Note: I had to reduce the number of
ellipses in order to get a smooth animation, so the quality is a bit lower than
the full visualization.
How To
Again, if you just want to create an image, check out the
code on Matteo’s GitHub repository. But, if you’d like to create it in Tableau,
here’s how.
1. If you’re familiar with Python, you can download the Python Code.
2. If not, you can download the executable package, PointArt.zip. Extract all the files to your computer, then run PointArt.exe. Sorry Mac users, but I think this will only
work for Windows, but if you really want to try this, I’m happy to help you get
Python set up on your machine.
3. The program will guide you through the process of selecting a source image
(needs to be jpeg format) and will then output three files: The csv and text
files noted above and the converted image.
4. Using the color text file, create your custom color palette.
5. Open Tableau and connect to the csv file.
6. The csv only contains data about each ellipse. In order to plot the points
around the perimeter of each ellipse, we’ll need additional data. For this, you’ll
need my Model data set, which contains the T values needed. Download the Model.xlsx
file, and add as a data source in Tableau.
7. Join the csv data to the model data using a join calculation (each side
should simply be “1”).
8. Create two parameters, Length Scale
and Width Scale. Both should be
float. These will be used to tune the lengths and widths of each ellipse. Start
out by setting Length Scale to 1.5
and Width Scale to 1.3.
9. Create a new calculated field called Point
X which will be the x coordinate of the points making up the perimeter of
each ellipse. The formula is as follows:
[X] + [Length]*[Length Scale]*COS([T])*COS(RADIANS([Angle])) - [Stroke Scale]*[Width Scale]*SIN([T])*SIN(RADIANS([Angle]))
10. Create another calculated field called Point Y which will be the y coordinate of the points making up the perimeter
of each ellipse. The formula is as follows:
[Y] + [Length]*[Length Scale]*COS([T])*SIN(RADIANS([Angle])) + [Stroke Scale]*[Width Scale]*SIN([T])*COS(RADIANS([Angle]))
11. Drag Point X to the Columns shelf
and Point Y to the Rows shelf.
12. Right click on each of them and change them to Dimensions.
13. Reverse the y axis.
14. Drag Ellipse ID to the Detail
card.
15. Change the mark type to Polygon.
16. Drag T to the Path card, then
change the pill to a Dimension. This will control the order in which each point
is connected to create a polygon.
17. Drag Color to the Color card.
18. Set the Color pill to be sorted automatically. This will make it easier
for us to apply our color palette.
19. Click on the Color card, select your custom color palette, and apply it.
20. Clear the sort on the Color
pill.
21. Using the Ellipse ID pill on
the detail card, make sure its sort is set to Data Source Order, which should be
in numeric order.
22. Drag the Ellipse ID pill up so
that it is above Color on the marks
card. This will set the z-order properly so that the we have the right overlap
in our ellipses.
23. If desired, fine-tune the size of the ellipses using the Length Scale and Width Scale parameters.
24. Make other adjustments as desired including gridlines, fixed axes, etc.
Or, if all those steps are too complex, feel free to
download my workbook and then simply swap out the data and color palette with
your own.
If you create
any pointillistic art using this method, please share with me. I’d love to see
what you create!! And don’t forget to check out Matteo Ronchetti’s article and
further work on his website, https://matteo.ronchetti.xyz.
You can also follow him on Twitter: @matteoronchetti
Ken Flerlage, May 25,
2018
No comments: