For recipe authors#
You want to author recipes by hand, sweep multiple axes, register custom models / preprocessors / target transformers, or use the YAML grammar deeply.
If you want to …#
Goal |
Page |
|---|---|
Read the layer-contract design that recipes encode |
|
Look up the data-task axes (source / target / predictor / horizon / …) |
|
Read the consolidated Layer 1 data-handling policies (axis-by-axis) |
|
Register a custom model / preprocessor / feature block / combiner / target transformer |
|
Quick reference for all three register_* APIs in one page |
|
Author a custom target transformer (fit-window, inverse, leakage rules) |
|
Run L1 / L2 / L3 in isolation to debug a recipe or iterate on a layer |
|
See package-level recipe defaults |
Working examples#
Bundled YAML recipes covering each layer live under examples/recipes/ in the
repo. The archive_v0/ subdirectory inside that tree holds older recipes that
are kept for replication only — start from the unprefixed files (e.g.
l4_minimal_ridge.yaml, l3_mccracken_ng_baseline.yaml).
For end-to-end replication walkthroughs see Replications.
Bridging audiences#
For the why of each layer (boundary contracts, sink names, manifest fields) read the canonical architecture pages.
For the public API the recipe ultimately drives, see
encyclopedia/public_api.md.To browse every recipe axis × option (with full OptionDoc prose) see the encyclopedia.