There videos for this part of the course are incomplete. We’ll cover CART in detail in the lectures.
Start with this post
and then move on to the sections below.
Worked example 1 - letter-to-sound
The first few steps in building a CART for letter-to-sound. Follow along using the provided data set.
The training data
We need a set of labelled training data, because this is supervised machine learning.
The root node
We start by placing all the training data at the root node, and calculating its entropy.
Trying the first split of the root node
Next, we try splitting the data at the root node using one of the available questions...
Other possible splits of the root node
...and repeat for all other available questions. The best one is placed in the tree. Then we recurse.
Worked example 2 - phrase breaks
Another worked example in which you have to decide what predictors to use, then build the tree.
The task
We want to predict the locations of phrase breaks, just from text. Our chosen method is a simple form of…
The training data
Supervised machine learning starts with training data, labelled with the value of the predictee. You now need to decide what…
The questions
To use your chosen predictors in a CART, you need to devise binary yes/no questions that query their values.
Label the test data
The tree can now be used to make predictions for unseen test data, where only the predictors' values are known.
Related forum
-
- Topic
- Voices
- Last Post
- You must be logged in to create new topics.
This is the new version. Still under construction.