blocks-transformer/README.md

14 lines
1.2 KiB
Markdown
Raw Normal View History

2025-12-04 10:50:08 -05:00
# PD V3 Pre-Processing
- **Inputs:** Raw Prod/Live feature values for all PD v3 models.
- **Outputs:** Three treated feature dictionaries (`model_a_features`, `model_b_features`, `model_t_features`).
- **Artifacts:** Driver CSVs stored under `drivers/`.
- **Tests:** `python -m unittest sequence-3.pd_v3_pre_processing.test_block`.
- **Signature:** `__main__` keeps an explicit typed parameter list covering every raw feature (int/float/str) and builds the record from those args before treatment; keep it aligned with `request_schema.json`.
## Schema notes
- `request_schema.json` now enumerates every raw Prod/Live variable as a top-level property (no `record` wrapper) so the contract mirrors `driver/request_schema_sample.json`.
- `response_schema.json` reflects the array-based payload returned by `block.__main__` (`model_*_features` are ordered lists of `{ "name": feature, "value": treated_value }` objects with feature name enums to enforce the allowlists).
- Both schemas are frozen and describe the flat/array-of-dict structures shown above; do not introduce dict-of-dicts or object-of-dicts in these files even if a future change reorients the Python implementation.