Interactive application for exploring single-objective Gaussian Process regression across different kernel functions.
This tool visualizes how Gaussian Processes model uncertainty over functions. Given a set of observed data points, the GPExplorer computes a posterior distribution over possible functions, providing both predictions and confidence intervals.
Key features:
- 8 kernel types (RBF, Matern, Periodic, Linear, Polynomial, etc.)
- Kernel composition via addition or multiplication
- Adjustable noise and hyperparameters
- Custom data input or CSV upload
- Posterior sampling visualization
- Covariance matrix diagnostics
pip install -r requirements.txtstreamlit run app.py| File | Description |
|---|---|
app.py |
Streamlit application |
kernels.py |
Kernel implementations and composition |
gpr.py |
Gaussian Process Regressor class |
requirements.txt |
Python dependencies |
| Kernel | Use Case |
|---|---|
| Squared Exponential (RBF) | Smooth, infinitely differentiable functions |
| Matern 3/2 | Once-differentiable, realistic for physical processes |
| Matern 5/2 | Twice-differentiable, common default choice |
| Rational Quadratic | Functions with varying lengthscales |
| Periodic | Seasonal or cyclic patterns |
| Linear | Linear trends |
| Polynomial | Polynomial relationships |
| Exponential | Rough, non-differentiable functions |
Kernels can be combined using + (additive) or × (multiplicative) operators.