Caching utilities that prevent data leakage by including split identifiers in cache keys. Uses memoise with cachem for efficient LRU caching.
Details
CRITICAL for zero-leakage: Cache keys MUST include: - Split/fold identifier (prevents cross-fold contamination) - Preprocessing parameters (ensures correct pipeline state) - Feature subset hash (for filter-specific caching)
This module ensures that cached filter scores or model artifacts from one CV fold are never reused in another fold.