Quickstart

This tutorial runs the smallest packaged ground-truth benchmark and opens the path to the generated report. It is intended as a first check that the library, CLI, estimators, result store, and reporter all work together.

Install

From a checkout of the repository:

pip install -e ".[reports]"

For a released package, use:

pip install "lrdbench[reports]"

Run the first benchmark

From the repository root:

python examples/quickstart_pure.py

The script runs configs/suites/smoke_ground_truth.yaml and prints:

• the run_id;
• the result store directory;
• the HTML report path;
• a command for validating the public output contract.

The installed package can run the same suite by name, without a repository checkout:

lrdbench run smoke_ground_truth

From a checkout, the manifest path also works:

lrdbench run configs/suites/smoke_ground_truth.yaml

Before fitting anything, preview the grid with --dry-run:

lrdbench run configs/suites/smoke_ground_truth.yaml --dry-run

The dry run materialises records and prints mode, n_records, n_estimators, n_fit_jobs, and global_seed without writing reports. For stress-test manifests it also prints the clean versus contaminated record counts. Use this check before launching larger public or custom suites.

Packaged suite names can also be inspected with:

lrdbench list-suites

Validate the output

Use the result_store path printed by the command:

lrdbench validate-output reports/<run_id>

The command returns 0 when required files and minimum CSV headers are present. It does not verify values or scientific completeness; see output validation.

What was benchmarked

The smoke suite generates one fractional Gaussian noise series with known target value and runs the bundled RS estimator. Because this is ground-truth mode, truth-based metrics such as bias, MAE, RMSE, empirical coverage, and interval width are meaningful.

For a fuller explanation of those metrics, continue with Ground-truth benchmark.