Getting Started¶
TrustLens is designed to be zero-friction. This guide will get you from zero to a production-grade model audit in less than two minutes.
Framework Support¶
TrustLens works out-of-the-box with common ML frameworks:
scikit-learn: All classifiers inheriting from
ClassifierMixin.XGBoost: Both
XGBClassifierand rawBoosterobjects.
The library automatically detects your framework. If you use a different framework or external inference system, you can still use TrustLens by providing predictions manually:
# Pass model=None if you only have results, not the model object
report = analyze(None, X, y, y_pred=my_preds, y_prob=my_probs)
[!NOTE] Degraded Mode: If you provide
y_predbut noty_prob, TrustLens enters “Degraded Mode”. It will skip confidence-based metrics (like calibration) while still auditing accuracy and fairness.
Installation¶
pip install trustlens
Minimal Working Example¶
The primary entry point is trustlens.analyze(). It orchestrates the entire evaluation pipeline and returns a TrustReport.
from trustlens import analyze
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
# 1. Prepare your model and data
X, y = make_classification(n_samples=1000, n_features=20, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
model = RandomForestClassifier().fit(X_train, y_train)
# 2. Run the decision-support audit
report = analyze(model, X_test, y_test)
# 3. Inspect the results
report.show() # Prints a professional summary to the console
What happened?¶
When you call analyze(), TrustLens performs a deep diagnostic sweep:
Calibration Check: It measures if your model’s 90% confidence actually means 90% accuracy.
Failure Modes: It identifies high-confidence mistakes (the “Confidently Wrong” pattern).
Trust Scoring: It aggregates all signals into a single Trust Score (0-100) and provides a deployment Verdict.
Next Steps¶
View the Features to understand the metrics.
Check the Use Cases for domain-specific examples.
Explore the API Reference for advanced configuration.