The Perfect Visualization

There are many rules about how to visualize data. We know how to encode specific types of data, what visual encodings work well, and what does not work so well. But is there such a thing as a perfect visualization for a given set of data?

This is a topic that comes up every now and then. In mathematics and some parts of computer science, an algorithm can be shown to be optimal. That requires a way to measure the outcome, though. Without a metric to optimize, there is no optimization. In algorithms, that is usually time, and sometimes error (for approximate or stochastic algorithms). The better algorithm is the one that produces the more desirable value in the metrics.

For visualization, the problem could be stated as: given a data set, what is the best visualization? And can you prove that this is the best? And are there multiple ways of visualizing the data that are equally good?

Visualization can be automated to an extent. We know how to show a given set of data based on the data types that are present. This is not new, much of this work was done in the 1980s. Do you have a categorical dimension and a continuous one? Bar chart. Time and a continues measure? Line chart. Etc.

It gets a bit more complicated than that when the number of data dimensions gets larger, of course. There is also a wide variety of chart types that are not [...]