Mike Godwin

MAT 259 - Data Visualization

My goal for this course was to produce a series of visualization experiments. I wanted to make quick “sketches” of the different topics that came up in our reading and discussions. This would allow me to familiarize myself further with the Processing programming language and to experience firsthand the associated problems and benefits of different methods of depicting information.

I recently read Michael Pollan's “Botany of Desire” and was so intrigued by the chapter on apples, that I decided to explore Plant Genetic Resources Unit division of the
Agricultural Research Service for possible data sets. A little sleuthing turned up the National Plant Germplasm System and a wealth of downloadable datasets.

It was at this point that “preprocessing” became the word of the month. In my case preprocessing meant taking an existing database that was fine tuned for MS-DOS circa 1990 and figuring out how to massage some sort of useful information from it.

Crop Common Names

My first visualization was an exploration of the database itself. What did all the acronyms mean, and how might these be visually interesting?

I became intrigued by the “common names records” for the various crops and created this visualization to investigate how common names were associated with the number of species that shared those common names.

Crop Diversity Comparison

My next sketch was an experiment with 3 dimensional representations of data. I chose three axes from the dataset and plotted each crop within those parameters.

While the exercise was helpful for me from a programming perspective, I quickly realized that reading information from this sort of image is very challenging and often vague.

The output highlights unusual crops, but obscures many details in the process. In my next sketches I decided to revisit 2D images in an attempt to recover some of the lost information.

Kohonen Self-organizing Color Map

This is an implementation of the standard first Kohonen programming project. The sketch generates 200 random colors and then uses a Kohonen algorithm to organize the points in a coherent “map.”

This is an easy way to see if the program is working correctly, but I didn't like how the actual values of the input data are not visible in the end result. In my final sketches I explored ways of solving this problem.

Kohonen SOM Experiments

Still concerned with the loss of information with generalized visualizations; I wanted to find a way to make the Kohonen algorithm's output more visibly connected to the original data points.

This visualization sorts 25 grays according to their red, green, and blue values (which are always equal for grays). This is mostly a verification that my program was working properly in a list format.

Kohonen Crop application

This is the concept of the layered Kohonen mapping visualization applied to a sample set of data from the Plant Genetic Resources Unit Germplasm database. The Kohonen algorithm sorts the list of crops into a list that is grouped according to the attributes shown along the right. This approach takes advantage of the Kohonen sort but makes each weight in the multi-dimensional map apparent.