P45: A Python C4.5 implementation

This is an accurate (or as accurate as possible) re-implementation of the classic C4.5 decision tree algorithm. The implementation includes:

separate-and-conquer recursive partitioning method
error-based pruning
copes with both numeric and categorial attributes directly
support for missing values

Current assumptions (when using the build_decision_tree function directly):

class attribute is the last attribute of the dataframe
data types on the dataframe are correctly set for each column: str for categoritcal attributes and float for numeric attributes

Running P45

There are two options to run the algorihtm, either directly from the command-line or imported as a library.

You may need to install additional dependencies to be able to run the algorithm — easiest way to do that is running the following in your terminal:
pip install -r requirements.txt

Command-line

The command-line allows you to run the algorithm and provide its parameters. The complete set of options can be viewed by running:

python p45.py

This will display a list of commands and their short description:

usage: p45.py [-h] [-m cases] [--seed seed] [--unpruned] [--csv] [-t <test file>] <training file>

P45: A Python C4.5 implementation.

positional arguments:
  <training file>  training file

optional arguments:
  -h, --help       show this help message and exit
  -m cases         minimum number of cases for at least two branches of a split
  --seed seed      random seed value
  --unpruned       disables pruning
  --csv            reads input as a CSV file
  -t <test file>   test file

The minimum set of parameters is to specify the training file:

python p45.py iris.arff

After a successful run, the algorithm produces a decision tree:

P45 [Release 1.0]                                      Thu April 7 06:00:00 2022
-----------------

    Options:
        Pruning=True
        Cases=2
        Seed=0

Class specified by attribute 'Class'

Read 150 cases (4 predictor attributes) from:
    -> iris.arff

Decision tree:

petal-width <= 0.6: Iris-setosa (50.0)
petal-width > 0.6: 
|    petal-width > 1.7: Iris-virginica (46.0/1.0)
|    petal-width <= 1.7: 
|    |    petal-length <= 4.9: Iris-versicolor (48.0/1.0)
|    |    petal-length > 4.9: 
|    |    |    petal-width <= 1.5: Iris-virginica (3.0)
|    |    |    petal-width > 1.5: Iris-versicolor (3.0/1.0)

Time: 2.7 secs

Import

You can also add P45 as a dependency to your project:

import p45

Then you can you can use the build_decision_tree function to create a decision tree. The return of this function is a Node object that represents the root node of the tree. From the root node, you can classify new instances by calling the predict or probabilities functions.

Limitations

Currently the code can take a while to run when using large datasets. The most likely reason for this is the (over-)use of dataframe operations to split the data during the recursive tree creation procedure.

febo/P45

P45: A Python C4.5 implementation

Running P45

Command-line

Import

Limitations