trust-free 0.9.3

Transparent, Robust & Ultra-Sparse Trees (TRUST) - Free Version

trust-free

Transparent, Robust & Ultra-Sparse Trees (TRUST) - Free Version

This package provides the free version* of our TRUST algorithm, a SOTA interpretable machine learning model.

Currently designed only for regression tasks. Future releases will also tackle other tasks e.g. classification.

*The free version is slower and is limited to datasets of at most 5,000 rows and 20 columns. It does not include premium functionality like:

• Large Language Model (LLM) integration for enhanced explanations
• Signed (+/-) variable importance plots
• Interaction ALE plots
• Prediction confidence intervals
• Out-Of-Distribution detection

Installation

You can install this package using pip:

pip install trust-free

Note: This package includes a precompiled binary and is currently only compatible with macOS 11+ on ARM64 architecture.

Usage

Here are two basic examples of how to use the TRUST algorithm:

from trust import TRUST # note the import name is trust, not trust-free
from sklearn.datasets import make_regression
from sklearn.model_selection import train_test_split
from sklearn.metrics import r2_score, mean_squared_error

# Example 1: Generate synthetic sparse regression data
X, y, coefs = make_regression(n_samples=5000, n_features=20, n_informative=10, coef=True, noise=0.1, random_state=123)
print(coefs)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=123)
# Instantiate and fit your model
model = TRUST()
model.fit(X_train, y_train)
# Predict and print results
y_pred = model.predict(X_test)
print("Predictions:", y_pred[:5])
print("True y values:", y_test[:5])
print("test R^2:", r2_score(y_test, y_pred))
# Obtain prediction explanation for first observation
model.explain(X_test[0,:], y_pred[0], actual=y_test[0]) 
# Obtain (conditional) variable importance by Ghost method (Delicado and Pena, 2023)
model.varImp(X_test, y_test, model, corAnalysis=True)

# Example 2: the famous diabetes dataset (n=442, p=10)
import pandas as pd
from sklearn import datasets

Diabetes = pd.DataFrame(datasets.load_diabetes().data)
Diabetes.columns = datasets.load_diabetes().feature_names
diab_target = datasets.load_diabetes().target
Diabetes.insert(len(Diabetes.columns), "Disease_marker", diab_target)
Diabetes_X = Diabetes.iloc[:,:-1]
Diabetes_y = Diabetes.iloc[:,-1]
RLT_Diabetes = TRUST(max_depth=1)
RLT_Diabetes.fit(Diabetes_X,Diabetes_y)
y_pred_TRUST = RLT_Diabetes.predict(Diabetes_X)
# Tree plotting requires Graphviz to be installed in your system path
# You can use e.g. Homebrew: brew install graphviz or Conda: conda install -c conda-forge graphviz
RLT_Diabetes.plot_tree("Diabetes") #will save "tree_plot_Diabetes.png" in your working directory
# Obtain prediction explanation for first observation
RLT_Diabetes.explain(Diabetes_X.iloc[0,:], y_pred_TRUST[0], actual=Diabetes_y.to_list()[0])
# Obtain variable importance with 2 different methods: Ghost and permutation
RLT_Diabetes.varImp(Diabetes_X, Diabetes_y, RLT_Diabetes, corAnalysis=True) #Ghost method
RLT_Diabetes.varImpPerm(Diabetes_X, Diabetes_y, RLT_Diabetes) #Permutation method

License

This software is provided under a Proprietary - Permissive Binary Only license. For detailed terms, please refer to the LICENSE file included with the distribution or visit our official website below.

More Information

For more details, documentation, and information about the full (paid) version of the TRUST algorithm, please visit our official website:

https://adc-trust-ai.github.io/trust/

Further details can be found in our preprint on arxiv:

https://www.arxiv.org/abs/2506.15791