Benchmarking

Public Exports

uncertainty_flow.benchmarking exports:

• BenchmarkConfig
• BenchmarkResult
• BenchmarkRunner
• AVAILABLE_DATASETS
• CHRONOS_DATASETS
• list_datasets
• list_datasets_by_domain
• load_dataset
• download_dataset
• TuningResult
• auto_tune

Module Structure

• flow.py: BenchmarkFlow orchestration module
• providers.py: provider interface seam and stable built-in model providers
• configs.py: benchmark/build configuration contracts
• results.py: benchmark result contracts
• sinks.py: output serialization policy via ResultSink
• runner.py: public compatibility adapter over flow/providers/sinks

Stable Model-Name Contract

Built-in benchmark names remain stable:

• quantile-forest
• conformal-regressor
• conformal-forecaster

Lifecycle

BenchmarkFlow lifecycle:

1. load
2. split
3. tune-per-run-context
4. fit/predict
5. evaluate
6. sink