For my next project at Berkeley, I cre­ated a real time visu­al­iza­tion of esti­mated train arrival times within the BART sys­tem. So next time you need to head over to the East Bay, just check the visu­al­iza­tion and you can see how far away your train is from the station.

More detail on the project and process behind it are doc­u­mented here. Thanks to BART for mak­ing their arrivals data available!

Launch project >

Yesterday marked the sec­ond day of the course I’m tak­ing at UC Berkeley titled, sim­ply, “Visualization.” I’ll be record­ing my most valu­able learn­ings from the course here on the blog, mostly for the ben­e­fit of fel­low DMIers. If you find this inter­est­ing, just fol­low along on the blog, or find more infor­ma­tion on the class wiki. Yes, every­thing related to the course is on there — the read­ings, com­pleted assign­ments, every­thing.

One of our read­ings for class was “The Eyes Have It: A Task by Data Type Taxonomy for Information Visualizations,” by Ben Shneiderman. In it, Schneiderman out­lines a tax­on­omy that cov­ers nearly all types of data:

• 1D (point, one value)
• 2D (x/y, planes, maps, etc.)
• 3D (x/y/z, space, objects with depth)
• Temporal (4 dimensions)
• Multi-dimensional (or nD, basi­cally any datum with >4 dimen­sions or values)
• Trees (hier­ar­chi­cally struc­tured data)
• Networks (graph struc­ture, or non-hierarchical, but related)

I rec­om­mend spend­ing a few min­utes with his article.

We also talked about Nominal, Ordinal, and Quantitative data.

• Nominal data is named, or labeled (e.g. apples, oranges, starfish, race­cars), although the labels are not related in any direct way.
• Ordinal data is ordered (e.g. grade A meat, grade AA, and grade AAA). The order is known—whether from low to high or best to worst — but not the rel­a­tive dis­tance between each mea­sure. (For exam­ple, we know that grade AAA meat is dif­fer­ent from grade AA, but we can’t say whether it’s 2.5 times bet­ter or 88.3 times worse than AA.)
• Quantitative data is quani­ti­fi­able in that its rel­a­tive posi­tion to other data can be eas­ily iden­ti­fied. (22 is 12 more than 10.)

Whether you use N, O, or Q depends on what your goal is for the visu­al­iza­tion. For exam­ple, you could start with the num­bers 10.5, 24.8, and -7.1. Your con­cep­tual model tells you what the mean­ing of these val­ues is. For exam­ple, the con­cep­tual model could be dis­tances, angles, or tem­per­a­tures. Let’s use temperatures.

• If con­sid­ered nom­i­nally, the val­ues could be cat­e­go­rized as burned or frozen.
• If con­sid­ered ordi­nally, the val­ues could be cat­e­go­rized as cold, cool, warm, or hot.
• If con­sid­ered quan­ti­ta­tively, the val­ues would rep­re­sent a range of rel­a­tive tem­per­a­ture values.

Why bother with all this? Because not all data types can be visu­al­ized in all ways. Jacques Bertin was the first to approach this sub­ject in depth, with his 1967 book Semiology of Graphics. In it, he inden­ti­fied attrib­utes of visual lan­guage (e.g. posi­tion, size, shape) that could cor­re­spond to the dif­fer­ent types of infor­ma­tion com­mu­ni­cated. Nominal data is more eas­ily rep­re­sented than quan­ti­ta­tive data, it turns out. For exam­ple, it’s dif­fi­cult to sequence using color (since the eye does not nat­u­rally per­ceive blue com­ing after yel­low or before red), although color is great for label­ing nom­i­nal val­ues (cities, sub­urbs, dis­puted territories).