You'll notice the packages mention fractals, of which there are a few in there, but lately I've been using this project to get into some neural network stuff. I devised a few programs to train neural networks on some basic geometrical problems.
The first problem I devised involved trying to see if a simple neural network could tell the difference between a set of circles where there was room for one more and a set where there was not. I could not get it to train reliably, which I found interesting.
Just to make sure I was doing anything right at all, I created a program to train a neural network to tell me if two circles overlap. That program had no trouble creating a very high degree of confidence after a dozen rounds or so. Given this result, I'm left to consider what aspects of the first problem lend themselves less to the particular solution I tried.
Anyway, that's the news from lake Wobegon.
