Chapter 5 Summary

5.1 Classification Problem

What are some statistical properties of two families of classifiers built from data?

5.1.1 Similarity Classifiers

5.1.2 Emprirical Risk Minimization

5.2 No Free Lunch Theorem

“Pointwise Statement”:

“Uniform Statement”:

Example of Pointwise Statement vs Uniform Statement:

No Free Lunch Theorem 1:

Practical Corollary:

No Free Lunch Theorem 2:

Practical Corollary:

Observation: we can not quantify the universal convergence rate.

Conclusion: No magic classifier that suits every situation.