Photographic Timelines

[DrQ]

Here's an interesting alternative view of the world. But what is it and is it good for anything besides art?

As explained in this posting on SingularityHUB, a conventional camera has a short exposure time with a large (i.e., widish) field of view. The above image is formed in a complementary way: a narrow field of view (window) over a long period of time. In other words, think of time as running along the bottom of the image to form a time axis with the distant past to the left and the recent past to the right. In addition, imagine the time axis as being divided into equal intervals or windows showing a different view in each time-slice. But that's the same thing we see in a conventional time-series of performance data. The width of each window corresponds to the sample period.

To me, this reinforces my view that time-series are one of the worst ways to analyze data; despite the fact that it is still the most common paradigm for presenting performance metrics. Our brain simply is not wired to accommodate it. It's a bad cognitive impedance match. Indeed, it has been discovered that our visual cortex receives pre-processed signals from our retina that are framed in time as images, very similar to movie frames. This phenomenon is cited in our 2007 CMG paper. In that sense, animation would seem to be a better paradigm that these Photographic Timelines.

From the standpoint of performance analysis, apart from seeing any obvious pattern of spikes in time-series data, which certainly can provide a useful starting point, you don't get very far. In fact, one of the things that held me up from posting to this PerfViz group yesterday was the my uncertainty over why the cars appear shortened; like they've undergone a Lorentz contraction. It looks completely unintuitive and is not totally explained by the time-series analogy. You have to read the text of the Singularity post carefully to realize that the artist mucks with the images (quite apart from rendering them as paintings). It seems he probably contracts the cars manually to fit into his selected spatial widow-slice and still have them be recognizable as a car.

So, quite the contrary to offering "a brilliant way to teach physics or philosophy," I think Photographic Timelines could be a good way to demonstrate why time-series are already a counterintuitive way to present data. What do you think?