Wednesday, July 25, 2007
The first few graphs I made just by futzing around with Emacs and gnuplot. When I had to more processing by smoothing histograms and such I decided to use Scheme to manipulate the data. When I started thinking I had to convolve the data with gaussians I decided that I should use MIT Scheme because it has a pretty good compiler. The script that probes the machines was written in Scsh, and I made sure to emit the samples in a trivially readable format. Scheme has been great for experimenting with this. I've written a fair amount of ad-hoc code for generating the data for gnuplot, and most of the resulting graphs are uninformative. When I find some presentation that does show something of interest, I just replace the specific literals with variables and wrap a lambda around it. I only mention this because I was surprised at how easy it was to transition from a small collection of random fragments of code to a simple toolkit by mixing and matching higher-order abstractions and by noticing common patterns. It is completely undisciplined programming, but very powerful.
Posted by Joe Marshall at 12:16 PM