3.8. Extension: Interpreting The Output Graph

Let’s see how the training samples ‘flow’ through our classification tree.

Let’s now focus in on one of the nodes.

• Ear Length (cm) <= 6.5 is the decision associated with the node.

• samples = 5 means that 5 samples have ‘flowed’ through this node. This corresponds to the 5 lines passing through the node in our visualisation.

• value = [1, 4, 0] tells you that there were 1 cat, 4 dogs and 0 rabbits that ‘flowed’ through this node. Again, you can tell by the colours of the lines. There is one yellow line for the cat and 4 purple lines for the dogs. The order of the list corresponds to the order of the class labels. In our case we had 1: cat, 4: dog, 0:rabbit.

• class = Dog means that the majority of the samples passing through this node belong to the dog class. In instances where there is a tie between classes, the class associated with the lowest number will be selected. E.g. [4, 4, 0] has equal cats (0) and dogs (1), so we default to cats.

Code Challenge: Extension: Building a Classification Tree

You have been provided with a csv file called astronomy.csv with data sourced from Wikipedia. This data contains the following columns:

• Name

• Mass (Earths)

• Density (g/cm^3)

• Class

The Class column values are, 0: Moon, 1: Planet, 2: Dwarf planet.

We will use this data to build a classification tree that can classify astronomical objects as a moon, planet or dwarf planet.

Instructions

1. Using pandas, read the file astronomy.csv into a DataFrame

2. Extract the 'Mass (Earths)' and 'Density (g/cm^3)' columns into the variable x

3. Extract the 'Class' column into the variable y

4. Convert both x and y to numpy arrays

5. Using sklearn, create a DecisionTreeClassifier model to fit to the training data, set the max_depth to 3

6. Export the tree using export_graphviz and set rounded=True and impurity=False

7. Save the tree as a png file.

Your figure should look like this:

Solution

Solution is locked