Tutorial notebooks

Five hands-on tutorials, each a Jupyter notebook in the repo under examples/. The pages here are rendered from the notebooks’ committed, executed outputs — what you see is what ran against a real PostgreSQL. Follow them in order:

#	Tutorial	Difficulty	Scenario
1	Basic aggregations	Beginner	E-commerce: Customers → Orders
2	Temporal joins	Intermediate	Healthcare: Patients → Care plans
3	Deep nesting	Intermediate	Retail supply chain, depth 3
4	Custom primitives	Advanced	Financial analytics: Accounts → Transactions
5	Categoricals, output & imputation	Advanced	Food inspections: Facilities → Inspections

What each one teaches

1. Basic aggregations — parent-child aggregations (count, sum, mean, min, max, stddev, nunique), time windows (P7D, P30D), feature naming. Start here (it is the notebook form of the walkthrough).
2. Temporal joins — as-of join semantics, grace periods, point-in-time generation, the LEFT JOIN LATERAL SQL, and rolling stats.
3. Deep nesting — multi-level relationships (max_depth: 3), feature propagation across chains, and how the CTE structure grows.
4. Custom primitives — writing and registering your own aggregations (range, p95) and transformations (log1p, zscore, bin), then selecting them in config.yaml.
5. Categoricals, output & imputation — fixed-vocabulary one-hot encoding (role: categorical / role: identifier), the feature manifest, output formats (to_dataframe / to_arrow), and the imputation contract. Unlike 1–4, this one executes against PostgreSQL throughout.

To run them yourself:

git clone https://github.com/ccd-ia/featurizer.git && cd featurizer
uv sync && just db-up
uv run jupyter lab examples/