arboreal

Python package to modularly create decision trees. arboreal uses pandas extensively to build decision trees.

Some notable features of arboreal:

1. First-class missing value support. Each decision tree node splits three ways: left, right, and missing. You don't have to impute the missing values. Missing values are handled in first class.

2. First-class categorical feature support. Categorical features are not converted into one-hot encoding. arboreal splits a tree node by treating categorical features as-is. You don't need to convert categorical features to a one-hot feature set.

3. Modular tree building. You can build your own tree manually, if you wish. You can build a part of the tree automatically using data and build the another part of the tree manually be specifying feature and split points to split on.

This is a work in progress. Stay tuned!

Installation

arboreal can be installed using pip

pip install --upgrade arboreal

Try it out

import numpy as np
import pandas as pd
from arboreal import Node, Tree, split_ordered_manual

# Build a small example dataset
data = pd.DataFrame({'feature1': [1, np.nan, 3, 4], 'feature2': ['a', 'b', 'c', 'd'],
                     'class': [True, False, False, False]})

# Initialize a tree
t = Tree(data=data, features=['feature1', 'feature2'], label='class', name='example')
print(t.root)
print(t.root.children())  # no children yet


# Manually create a split for 'feature1' at split_point = 3
split = split_ordered_manual(data=t.root.data, feature='feature1', label='class', split_point=3)
t.root.perform_split(split)

# Check the result
print(t.root.children())  # had three children: left, right, and missing

# Predict at any node
print(t.root.predict())
print(t.root.left.predict())
print(t.root.right.predict())
print(t.root.missing.predict())