lime-ndt 0.1.3

Enhancing LIME through Neural Decision Trees

lime_ndt

lime_ndt is a Python library that introduces an enhanced version of the LIME technique for model explainability, leveraging Neural Decision Trees (NDTs) for improved local and global interpretability of machine learning models.

Features

• Enhanced LIME explanations using Neural Decision Trees.
• Support for tabular data.
• Tools for extracting, analyzing, and comparing decision trees and forests.
• Utilities for submodular pick and discretization.
• Integration with Keras and scikit-learn.

Example Usage

Below is a minimal example showing how to use lime_ndt to explain a regression model on the diabetes dataset:

from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from lime_ndt.lime_tabular import LimeTabularExplainer
from lime_ndt.utils.ndt_sklearn_wrapper import NDTRegressorWrapper
import matplotlib.pyplot as plt

# Load the diabetes dataset
diabetes = load_diabetes()
X = diabetes.data
y = diabetes.target

# Split into train/test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)

# Train a Random Forest regressor
rf = RandomForestRegressor(random_state=42)
rf.fit(X_train, y_train)

# Prediction function
def predict_fn(X):
    return rf.predict(X)

# Create the LIME explainer
explainer = LimeTabularExplainer(
    X_train,
    feature_names=diabetes.feature_names,
    class_names=None,
    discretize_continuous=True,
    mode='regression',
)

# Create the local NDT model
model_regressor = NDTRegressorWrapper(D=X_train.shape[1])  # gamma=[1,100] , max_depth = 5

# Explain a test instance
exp = explainer.explain_instance(
    X_test[8],
    predict_fn,
    num_features=10,
    model_regressor=model_regressor
)

# Visualize the explanation
exp.as_pyplot_figure()
plt.show()

Requirements

• Python 3.7+
• numpy, pandas, scikit-learn, keras, tensorflow, matplotlib

See requirements.txt for the full list.

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.