C4.5 Decision Tree

Introducion

An implement of the classic ML algorithm C4.5 decision tree.

Quick Start

First to import the c45 modules, and initialise.

from c45 import MetricGenerator
from c45 import DecisionTree

tree = DecisionTree()

To apply the decision tree on your own dataset, you need to rigister metric functions into an object of MetricGenerator class.

There are some templates in MetricGenerator, for nominal, ordinal, and contineous feature attributes. You can easily create a set of binary classify metrics and automatically register it. Besides, it's also possible to code your own metric functions and register them into the generator.

Here is am example:

feature = "color"
values_set = ["red", "blue", "green"]
metrics_set = MetricGenerator.nominal_template(feature,values_set)
tree.metric_generator.register(feature, metrics_set)

read data into the root node of the tree object

FILE_PATH = "dataset.csv"
tree.read_data(FILE_PATH)

train the tree

tree.train()

run on the testing set

TEST_FILE_PATH = "testset.csv"
precision = tree.run_test(TEST_FILE_PATH)
print("error rate on test set: ", 100 * (1 - precision), "%")

Dependencies

The c45 module actually only need numpy and python3 to run.

1. numpy
2. python 3

In the example main.py, I used pandas and sklearn to split data into training set and testing set. However, they are not necessary for the algorithm. You can use native methods to process data set.

Example

There is an example on iris data set. You can check it in main.py.

Roadmap

• Visualise the decision tree.
• pruning algorithm.