Train Value HPM With Playoffs

This HIPSTER PM variant adds each completed historical season's playoff stints to that season's regular-season fit. Those outcomes therefore update the state used for the following season. The target season's regular season and playoffs remain frozen evaluation cohorts and are never used to form their own forecast state.

Only seasons with validated playoff stints are augmented. The resulting season_model_metadata.parquet records the exact coverage, including every season that remains regular-season-only because its playoff stints have not yet been reconstructed to the same contract.

uv run nba-train-forward-centered-value-conditioned-aging-playoffs-bounded-hierarchical-portable-matchup-contextual-rapm \
  --through-season 2025-26

The model retains the value-conditioned aging prior, exposure-gated cold starts, bounded hierarchical P-splines, and attributable portable-matchup context from the regular-only HPM.

Outputs

artifacts/models/forward_centered_value_conditioned_aging_playoffs_bounded_hierarchical_portable_matchup_contextual_rapm/2025-26/

season_model_metadata.parquet records regular-season stint counts, included playoff games, excluded playoff games, and the total fitting rows for every season. The target row has zero included playoff games by design.