We've started analysing the curves from Team 2's iPads, by writing a computer program (in a language called MATLAB). Here's part of it:
Today I spent some time analysing the curves from Mutating Messages on the computer.
First of all I looked at the way the curves mutated in the experiment. I managed to write a computer program that makes curves mutate in a similar way. The program was based on something we mentioned in the video: Gaussian Processes. This gave me a statistical model for the curves.
I then just took the three final curves, and used the computer to work out what the network was. In our video this would be like only seeing the curves appearing on the last three desks and then trying to work out the arrangement of the remaining desks in the class. (In terms of evolution, I was trying to work out what the most likely evolutionary tree was, having seen only the leaves of the evolutionary tree).
I was really surprised how convincing the results were - the computer got it right, and was very sure of its answer. How can a computer be sure? It produces a number called the likelihood, which indicates how positive the results are. The likelihood was huge when I tried using the correct arrangement of desks (and not so big when I put in other arrangements). The results were just as good when I tried some data which I just made up and inserted by hand. Now I'm ready to put the real curves from Mutating Messages into the program!