Example Usage

This guide demonstrates how to use the obliquetree library’s Classifier for training decision trees with oblique splits and visualizing the results. The examples explore different configurations, including oblique and non-oblique splits, and highlight the library’s support for categorical data.

Imports

from obliquetree import Classifier
from obliquetree.utils import visualize_tree

from sklearn.datasets import fetch_openml
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, LabelEncoder
from sklearn.metrics import accuracy_score

---

Data Preparation

We use the “Adult” dataset from OpenML, apply preprocessing, and split it into training and testing sets.

RANDOM_STATE = 42

X, y = fetch_openml("adult", version=2, return_X_y=True)
y = (y == ">50K").astype(int)

cat_column = X.select_dtypes(exclude="number").columns
numeric_columns = X.select_dtypes(include="number").columns

encoder = LabelEncoder()
for col in cat_column:
    X[col] = X[col].astype(str)
    X[col] = encoder.fit_transform(X.loc[:, col]).astype(float)

X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=RANDOM_STATE)

scaler = StandardScaler()
X_train[numeric_columns] = scaler.fit_transform(X_train[numeric_columns])
X_test[numeric_columns] = scaler.transform(X_test[numeric_columns])

cat_idx = [X_train.columns.get_loc(col) for col in cat_column]

Note

Scaling is important for oblique splits. Since obliquetree supports categorical data, we pass the indices of categorical columns via cat_idx.

---

Example 1: Oblique Tree with Categorical Support

clf = Classifier(
    use_oblique=True,
    max_depth=2,
    n_pair=2,
    random_state=RANDOM_STATE,
    categories=cat_idx,
)

clf.fit(X_train, y_train)

train_pred = clf.predict(X_train)
train_mse = accuracy_score(y_train, train_pred)

test_pred = clf.predict(X_test)
test_mse = accuracy_score(y_test, test_pred)

print(train_mse, test_mse)

Output:

0.8395075209521989 0.8513635246908525

Note

Here, n_pair=2 by default, meaning oblique splits are computed using only two features.

visualize_tree(clf)

---

Example 2: Axis-Aligned Tree with Categorical Support

clf = Classifier(
    use_oblique=False,
    max_depth=2,
    categories=cat_idx,
)

clf.fit(X_train, y_train)

train_pred = clf.predict(X_train)
train_mse = accuracy_score(y_train, train_pred)

test_pred = clf.predict(X_test)
test_mse = accuracy_score(y_test, test_pred)

print(train_mse, test_mse)

Output:

0.826704157680653 0.8381786913438702

Note

This configuration removes oblique splits but still supports categorical data. Random state is not required for axis-aligned splits.

visualize_tree(clf)

---

Example 3: Axis-Aligned Tree Without Categorical Support

clf = Classifier(
    use_oblique=False,
    max_depth=2,
)

clf.fit(X_train, y_train)

train_pred = clf.predict(X_train)
train_mse = accuracy_score(y_train, train_pred)

test_pred = clf.predict(X_test)
test_mse = accuracy_score(y_test, test_pred)

print(train_mse, test_mse)

Output:

0.8238923316316781 0.8340021292277455

Note

In this configuration, both oblique splits and categorical support are disabled, yielding results consistent with standard scikit-learn trees.

visualize_tree(clf)

---

Example 4: Oblique Tree with n_pair Set to Total Features

clf = Classifier(
    use_oblique=True,
    max_depth=2,
    random_state=RANDOM_STATE,
    n_pair=X.shape[1],
    categories=cat_idx,
)

clf.fit(X_train, y_train)

train_pred = clf.predict(X_train)
train_mse = accuracy_score(y_train, train_pred)

test_pred = clf.predict(X_test)
test_mse = accuracy_score(y_test, test_pred)

print(train_mse, test_mse)

Output:

0.84021730228495 0.8489886168208992

UserWarning: Total features: 14, categorical features: 8. n_pair (14) exceeds the usable features, adjusting n_pair to 6.

Important

When n_pair exceeds the number of usable features, the library automatically adjusts it. Oblique splits do not support categorical features directly, so the n_pair count is reduced.

visualize_tree(clf)