When I first started working in statistics and data processing algorithms, or “Machine Learning” which has become the general term for a broad collection of methods like this, I found it difficult to navigate a seemingly disjointed literature. Literature was either very theoretical, such as laboriously proving convergence criteria for different learning methods, or applied to data without much discussion of underlying method.
Uniting the two approaches was seldom seen; I wanted a description of the methods being used in detail, with some helpful literature if necessary, but applied in a simple way using simple syntax in something like Python. It was hard to find this, so I made it myself after working it out the hard way in my head.
All modern “Machine Learning” methods are about drawing useful inferences from data, with some underlying model and usually an optimisation method. The latter two points are very important, and here I tackle simple examples that are not based on the big scary method: Neural Networks. There is a plethora of extremely good introductions to this from people far better than me at it, so I won’t tackle it here. I will just tackle what I know: Bayesian Statistics, and numerical optimisation techniques, and applying them to data (“Machine Learning”).
This is a collected into a loosely structured cookbook of methods anyone can use here: https://github.com/KentJames/Statistics-Cookbook