W07. State 04
Today we will learn about classifiers, using line-tracking as an example. Our modern world uses classifiers as the main detection paradigm because creating a simple "rule" for detectors is difficult if not impossible for most tasks with complex and real-world data. We'll see an example of pre-built classifiers that you can use with your robots, and start to think about how to design a system that can use a classifier.
Pre-readings and Videos
The readings for today are mostly here to make the connection between Bayes and classifiers. In reality, the Bayes classifier is just one of many.
Naive Bayes classifier
The Naive Bayes Classifier should look pretty familiar to you now that we've been talking about Bayes for two weeks. This article goes over the math that we won't cover in class.
Naive Bayes Video
A video outline of the math behind the Naive Bayes classifier.
Random Forest Classifier
There are many different kinds of classifiers. Random forest is a good one, and relatively simple to understand. This video is provided as a counterpoint to the above Bayes content, mostly to illustrate that different classifiers exist.
Summary of the Day
- Activity. Line following cases.
- Activity. Deep learning detectors in Processing.
- Activity. Labelling protocol for your robot.
- Class notes. Available here
Learning Goals
- Be able to articulate why simple rule-based detectors can often fail.
- Outline the basic goals and design parameters of classifiers.
- Design a labelling protocol for your robot.