Minimal recipe that trains a recommender from a local gzip-compressed CSV shipped alongside the recipe. No network access required.
recipe.yaml— uses local pathexamples/csv-local/interactions.csv.gzinteractions.csv.gz— small synthetic dataset (user_id,item_id,timestamp)
From the repository root:
mkdir -p artifacts
uv run recotem train examples/csv-local/recipe.yamlThe artifact is written to ./artifacts/csv_local_example.<sha>.recotem
(the .<sha> suffix comes from versioning: append_sha).
- A
csvdata source with explicitdtypeoverrides for ID columns. time_usertrain/test split (requirestime_columnon the schema).- IALS, CosineKNN, and TopPop algorithms compared via Optuna ndcg@20.
Use this example when you want to develop offline, or to study the recipe schema without an HTTPS round-trip. The tutorial-purchase-log example is the runnable end-to-end walkthrough that the getting-started guide is built around.