Friday, February 11, 2011

This is Why They Make Comic Strips about Workplaces

There have been a series of meetings featuring different combinations of people: me, one of my bosses, a "computer guy", an undergrad, a biologist, and a (very busy) statistician. We are trying to understand a biological model that is implemented in closed-source Java code.

Here is the plan: My boss talks, briefly, to the busy statistician about the problem. The statistician suggests that we should use a fractional factorial design and tells us how to generate the design -- but not how to interpret the results after we've run it. He gives us a few lines of R code that will generate the design. My boss has directed the undergrad to make the list of variables that will go into the design and to pick the levels for them. (This is easy for the booleans and hard for the doubles.) I will then use the R software to create the design. (This literally involves typing in maybe four lines of R code that the statistician gave us.) The "computer guy" will then have the supercomputer do all the runs that are prescribed by the design. And then I'm not sure what happens next. We may need to find the busy statistician. My boss is expecting miracles out of this process.

My understanding is that we are doing all of this so that we can understand the software. We want to know what all of the various settings and command-line flags do and how they change the output of the software.

I suggested instead of having a half-dozen people involved in writing Franken-code and running it on a supercomputer and applying a powerful but perhaps inappropriate statistical technique that maybe we could just call the developer and ask.