ce-performance-tuning

What problem does it solve?

Configure CE caching, parallel execution, batch-size tuning, and FAST feature filtering per ADR-003 and ADR-004 for faster explanations on large datasets.

Core Features & Use Cases

• Caching (ADR-003): In-process LRU cache for calibrator results; enable via cache=True or CE_CACHE=1 env var; eviction; flush via explainer.flush_cache(); deterministic keys use namespace + version_tag + payload hash.
• Parallel execution (ADR-004): Auto strategy, force parallel or serial, configuration with ParallelConfig; key files reference: src/calibrated_explanations/parallel/parallel.py and src/calibrated_explanations/perf/parallel.py.
• Batch size and chunking: chunk explanations to control memory usage; instance_chunk_size; memory hints for auto strategy.
• FAST feature filtering: reduces per-instance feature space; top_k; strict observability; env var CE_FEATURE_FILTER; debug governance entries.
• Diagnostic workflow and constraints: baseline, profile, apply, measure; constraints about opt-in caching, memory usage, and potential feature drops.

Quick Start

Tune CE performance by enabling caching, configuring parallel execution, and applying FAST feature filtering to accelerate explanations.