The best way to learn is to do problems. So here are some problems I've worked on. I try not to use existing packages as to learn the theory. All solutions in python.
Problems I have yet to do
Reproduces the polynomial curve fitting example in Bishop, Ch 1. The outline of the task is generate
- generate a synthetic dataset of N points (x,t) for a known function y(x) with some level of noise.
- Solve the curve fitting gression problem using error function optimization
- Observe the problems of overfitting this method produces
- Introduce regularization to overcome overfitting
- Use Bayesian estimation to produce an interval estimation of the function y
Reference