lrdbench documentation

Welcome to the lrdbench documentation: a reproducible benchmark framework for long-range dependence estimators on synthetic (ground truth and stress-test) and observational series.

Hosted site

This book is built with MkDocs and the Material for MkDocs theme. The canonical hosted copy is lrdbench.readthedocs.io.

Design authority

The tracked architecture, object contracts, manifest rules, and release-stability expectations are in Design specification. A short root-level traceability note remains in SPECIFICATION.md.

Repository layout (quick)

• Runnable suites: configs/suites/ (YAML + small data).
• Python package: src/lrdbench/.
• Target tree sketch (aspirational package split): lrdbench_repo_schema.txt at the repo root.

Where to go next

• Quickstart tutorial — run the first benchmark and validate the output.
• Tutorials — ground-truth, stress-test, observational, and custom-estimator workflows.
• Installation — editable install, extras, local mkdocs serve.
• Benchmark protocol — manifest modes, execution block, outputs.
• Bundled estimators — registry names, families, and key parameters.
• Data-driven estimators — RF/SVR/CNN/LSTM baselines and manifest-level training.
• Interpretation semantics — uncertainty, leaderboard, and failure rules.
• Release candidate freeze — historical pre-1.0 review of public APIs, schemas, columns, and metric names.
• Public small outputs — expected artefacts for public-small suites.
• Migration notes — public-surface changes across releases.
• Citation guidance — software citation and benchmark metadata expectations.
• Estimator contract — BaseEstimator and EstimateResult.
• Estimator status — public interpretation status for bundled estimators.
• Architecture — how the orchestration pieces fit together.
• Python API — selected autodoc pages.
• Public release roadmap — phased alpha/beta/v1.0 plan.
• Governance and maintenance — compatibility, review, and release policy.