GPClarity: Interpretability Toolkit for Gaussian Processes
Make your Gaussian Process models transparent.
GPClarity provides human-readable diagnostics, uncertainty quantification, and model introspection tools that bridge the gap between black-box GP predictions and actionable insights.
Getting Started
API Reference
Features
Kernel Interpretation: Translate cryptic hyperparameters into natural language insights
Uncertainty Quantification: Go beyond variance — detect extrapolation, miscalibration, and epistemic vs. aleatoric uncertainty
Complexity Analysis: Measure model capacity, detect overfitting/underfitting, get actionable recommendations
Optimization Tracking: Monitor hyperparameter convergence with publication-ready visualizations
Data Influence: Identify high-leverage points and quantify leave-one-out effects
Why GPClarity?
Gaussian Processes are powerful but opaque. GPClarity answers questions like:
“Is my kernel too simple or too complex?”
“Where can I trust my model’s predictions?”
“Which data points drive my model’s behavior?”
“Has my optimization converged or is it stuck?”
Installation
Basic (analysis only):
pip install gpclarity
Full (with GPy/emukit):
pip install gpclarity[full]
Development:
pip install gpclarity[dev]