Tutorials

Each tutorial is a complete, runnable notebook. The website includes the outputs from an executed run, including plots and printed results.

Start with First posterior, then choose the workflow you need:

Tutorial	Task
First posterior	Condition on a small dataset and plot a prediction.
Learning a function from derivatives	Combine slopes and curvature, then anchor the function with one value.
Learning from averages	Infer function values and accumulated integrals from window averages.
Experimental design	Optimize a batch of observations to reduce uncertainty about a target.
Sequential observations	Reuse compiled inference and update a posterior as observations arrive.
Fitting hyperparameters	Fit kernel and observation-noise parameters with MLE and MAP.
Sampling from a GP posterior	Draw values at fixed locations, or sample a function to evaluate and differentiate.
Implementing a custom operator	Infer a scalar function from vector readings using a custom operator.

The Concepts pages explain gpex's model and terminology. The API reference covers individual methods and their options. Additional configurations, including Bayesian optimization and advanced fitting, are available as executable examples.

To run the notebooks from a checkout:

uv sync --group dev --extra examples
uv run jupyter lab docs/tutorials/first-posterior.ipynb

Use Restart Kernel and Run All Cells to reproduce a notebook. Each runs independently and includes its own data. Blue denotes the main posterior, orange denotes observations or initial proposals, and purple denotes a comparison or path. Shaded bands show pointwise 95% latent intervals.