Thoughtful Machine Learning with Python

Chapter 10. Improving Models and Data Extraction

How do you go about improving upon a simple machine learning algorithm such as Naive Bayesian Classifiers, SVMs, or really any method? That is what we will delve into in this chapter, by talking about four major ways of improving models:

• Feature selection

• Feature transformation

• Ensemble learning

• Bootstrapping

I’ll outline the benefits of each of these methods but in general they reduce entanglement, overcome the curse of dimensionality, and reduce correction cascades and sensitivity to data changes.

They each have certain pros and cons and should be used when there is a purpose behind it. Sometimes problems are so sufficiently complex that tweaking and improvement are warranted at this level, other times they are not. That is a judgment people must make depending on the business context.

Debate Club

I’m not sure if this is common throughout the world, but in the United States, debate club is a high school fixture. For those of you who haven’t heard of this, it’s a simple idea: high schoolers will take polarizing issues and debate their side. This serves as a great way for students who want to become lawyers to try out their skills arguing for a case.

The fascinating thing about this is just how rigorous and disciplined these kids are. Usually they study all kinds of facts to put together a dossier of important points to make. Sometimes they argue for a side they don’t agree with but they do so with conviction.

Why am ...