I took a different path

Finished the take-home programming question for Metrics. If I hadn't fluffed around so much this morning, it would have taken less than 24 hours. As it is, it's early, and I enjoyed it.

The solid line here is my first attempt at writing a nonparametric estimator to the data points. I think you'll agree it's a very bad fit.

This fit is better - in fact, perfect, which shouldn't have happened. I somehow managed to cancel out the randomness and estimated the dependent variable data using only the dependent variable data.

This is more like it! Not too jagged, but nonetheless a good fit. This was the signal that the first half of the question was done.

It's difficult to demonstrate pictorially what the other half was about, because it involved finding an optimal value for a parameter used in the first half - except the plug-in parameter value we'd been given turned out to be exactly the optimal one, so the picture didn't change. The best I can do is point out that my results were precisely what Dr Metrics told us to expect:

Holiday time now. Oh dear... it's a while since I had one of those...