Examples

Five notebooks, each runnable end to end. Outputs on this site are generated by mkdocs-jupyter at build time against the current codebase, so nothing here can quietly go stale.

Notebook	What it covers
DGP Families	Every DGP family, the effect of complexity, and how to read the difficulty metadata
Corruptors	Feature corruptors and label noise, MCAR/MAR/MNAR, chained pipelines
Sweeps & Suites	severity_sweep, difficulty_sweep, experiment_grid, BenchSuite
End-to-End Workflow	Generate, corrupt, sweep, serialize, reload, benchmark with sklearn
Mini AMLB Benchmark	An OpenML task next to synthbench data, and what the error floor adds

Every notebook keeps n_samples <= 500 so they finish quickly.

To run them locally:

pip install "synthbench[docs,neural,io]"
pip install torch --index-url https://download.pytorch.org/whl/cpu
jupyter lab