Notebooks

The Quarto site includes executed versions of the core Python notebooks and links to the heavier benchmark notebooks. For the primary performance results, use the dedicated Benchmarks page.

Rendered in the site

These notebooks are part of the Quarto build and are rendered as documentation pages:

• example ERM-style linear, logistic, and Poisson regression examples.

• banana Optimizer demonstration on Rosenbrock’s banana function.

• gmm Linear IV GMM, nonlinear logit-style GMM, and bootstrap-based inference.

• autodiff_mnl JAX autodiff for multinomial logit with pyensmallen as the optimizer backend.

• regularization_comparison Regularized estimator demo with pyensmallen versus scikit-learn and MSE path plots.

Source-only notebooks

These remain in the repository but are not part of the always-executed Quarto build:

• likelihood_comparison.ipynb Large benchmark notebook comparing pyensmallen against SciPy, statsmodels, CVXPY, and JAX.

• likelihood_comparison_estimators.ipynb Benchmark notebook using the estimator-class API.

• synth.ipynb Synthetic-control style balancing weights and simulation work.

• compare_r.ipynb R notebook, not part of the Python-backed docs build.

Build behavior

The Quarto documentation build executes the rendered notebooks above. This keeps the core tutorial material live and testable without forcing every expensive benchmark notebook into the site build.

Running notebooks locally

Install the full dependency set:

uv pip install --python .venv/bin/python --no-build-isolation -e ".[full]"

Then launch Jupyter:

.venv/bin/python -m jupyter lab