Gradient-based Entire Tree Optimization For Oblique Decision Tree

This repository has been restructured to offer a more organized and user-friendly interface. GET (Gradient-based Entire Tree) is designed to induce oblique decision trees by optimizing the entire tree structure via gradient-based optimization. It supports both regression and classification tasks. For detailed information on the algorithm, please refer to the study “Can a Single Tree Outperform an Entire Forest?”, available at https://arxiv.org/pdf/2411.17003.

Features in this version:

• GETRegressor(): An oblique regression tree with constant predictions.
• GETSubPolRegressor(): An oblique regression tree with constant predictions, enhanced with a subtree polishing strategy.

New features will be added in next versions, including:

• tree path-based interpretability
• Classification tree implementations like GETClassifier() and GETSubPolClassifier()

Package Dependencies

• scikit-learn 1.5.0
• numpy 1.26.4
• pandas 2.2.3
• h5py 3.13.0
• torch 2.0.0+

Package Installation

pip install get-oblique

Package Description

GETRegressor class: oblique regression tree with constant predictions.

• Parameters:
  • treeDepth (int, default=4): The depth of the regression tree.
  • epochNum (int, default=3000): Number of training epochs used during optimization.
  • startNum (int, default=10): Number of random initializations for the tree optimization process (This increases the chance of finding optimal solutions).
  • device (str, default='cpu'): The computation device to use: 'cpu' or 'cuda'. Set to 'cuda' to enable GPU acceleration.
• Methods:
  • fit(X, y):
    Train the model using gradient-based optimization. Automatically moves data to the specified device and converts to float tensors.
  • predict(X):
    Predicts target values based on trained tree structure.

GETSubPolRegressor class: oblique regression tree with constant predictions and subtree polish strategy.

• Parameters:
  • treeDepth (int, default=4): The depth of the regression tree.
  • epochNum (int, default=3000): Number of training epochs used during optimization.
  • startNum (int, default=10): Number of random initializations for the tree optimization process (This increases the chance of finding optimal solutions).
  • device (str, default='cpu'): The computation device to use: 'cpu' or 'cuda'. Set to 'cuda' to enable GPU acceleration.
• Methods:
  • fit(X, y):
    Train the model using gradient-based optimization and subtree polish strategy. Automatically moves data to the specified device and converts to float tensors.
  • predict(X):
    Predicts target values based on trained tree structure.

Usage Example

To use the GETRegressor class:

import numpy as np
from sklearn.datasets import fetch_california_housing
from sklearn.model_selection import train_test_split
from GET import GETRegressor

# Load and prepare dataset
data = fetch_california_housing()

# X, y can be either Numpy arrays or Pytorch tensors, in this case they are numpy arrays
X, y = data.data, data.target

# Split into train and test
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Initialize the model
model = GETRegressor()

# Fit the model
model.fit(X_train, y_train)

# Predict
y_pred = model.predict(X_test)

# Print sample predictions
print("First 10 predicted values:", y_pred[:10])

Others

If you encounter any errors or notice unexpected tree performance, please don't hesitate to contact us.

License

This repository is published under the terms of the GNU General Public License v3.0 .