Train HPM v2.2
HPM v2.2 retains the empirical rebound realization state from v2.1 and swaps its raw usage, turnover, and top-two usage summaries for a completed-season conditional-logit terminal-action allocation state. It trains entirely from cached curated regular-season data and makes no NBA API requests.
uv run nba-train-hpm-v22 --through-season 2025-26 \
2>&1 | tee artifacts/logs/train-hpm-v22-2025-26.log
Follow the recursive seasonal fit with:
tail -f artifacts/logs/train-hpm-v22-2025-26.log
Each immutable artifact stores both calibration contracts:
season_rebound_calibration_metadata.parquetseason_usage_allocation_metadata.parquet
The fitted allocation model itself is embedded in each season's serialized context model, so deployment needs no raw event data for arbitrary-lineup inference.