It’s also a good learning experience for me. I learned how to manipulate text files using cell. By applying this new technique to my work I bet I can increase the efficiency of my company’s product by 50%, at least.

Cris: Good observation. The key is that I do not consider the correlation of the the programs, but the correlation of the *signature* of the programs. If the signature does not extract enough features from the programs, the self-correlation will be low. The signature that I used appears to be sufficient for a majority of the given test subjects, and I did not intend to make it too complicated.

I think it is weird, though, that an entry wins that computes correlation this way. Correlation of one program with itself should always be very high, I’d expect. It’s funny to see some very low values along the diagonal of the correlation matrices.

The optimization contests were getting a little long in the tooth, especially with the same people competing (and winning) every time. Let’s do this visualization one again!