I think I am in Quantified Self heaven now. I’ve got HabitBull (Misc Habits and Scores on Brain and Braverman Tests), FitBit (Activity + Weight + Food + Sleep tracking) , Timesheet (Work Project Time Tracking), Rescue Time (Computer Use Time Tracking), Fit Notes (Weight Training Tracking) all hooked into advanced machine learning analysis via a bunch of python scripts utilizing numpy and sklearn.
I can ask “what kinds of things that I did/ate/took/spent time doing yesterday are highly correlated with me losing weight and/or body fat?” and get a linear regression on that and a p-value. Currently, the most highly correlated with losing weight right now is calories eaten the previous day, but there are other significantly correlated stats (p-value < 0.05) that are more interesting. For example, the more time I spend doing sedentary activity, like software development, the less weight I lose.
I can find out which supplements influenced various brain test scores on the same day. I can blind myself with randomly numbered containers and then go back in and fill in what supplements I took retroactively. I can try all kinds of totally random things and see what they do and it’s all automatically tracked and analyzed.
My analysis takes conditional independence as assumed as do all naive bayes techniques. It also takes linearity as an assumption too, which is a weakness. To get around that, I could build a neural net and then use a genetic algorithm against it to generate ideas for non-conditionally independent relationships among things I’m doing.
More on this in the near future…