The Master Algorithm

Too difficult to follow for a non- data scientist. You can't explain this stuff without a lot of diagrams, videos etc.

This seems to be the only up-to-date book out there giving an overview of the discipline of Machine Learning, but nobody seems to be quite happy with it, and I can see why.

Domingos goes in detail on what he calls the “five tribes” of machine learning:

- Symbolism / Logic with it's decision trees and inverse deduction
- Connectionism with its multilayer perceptrons and backpropagation
- Evolutionaries with their genetic algorithms
- Bayesians with probabilistic inference
- Analogizers with their support vector machines

The level of complexity of his explanations and examples isn't well balanced, some are easy to follow, while other's are just too high-level and would require more hand-holding. Nevertheless you get a decent overview of the field.

The book fails where the author tries to insert himself, his opinions and his quest for the “Master Algorithm”. Or when he tries to add creative analogies, as when he describes the 5 machine learning strategies as boroughs of a city. And then spends multiple pages riffing on that analogy.