Expand description
Reproducible single-case reconstruction benchmarks.
The runner is generic over the existing reconstruction and measurement traits. Calling it once per concrete algorithm avoids a second algorithm registry or trait-object hierarchy.
Structs§
- Benchmark
Frame Record - Per-acquisition-frame benchmark identity and residual metric.
- Benchmark
Profile - Stable metadata for a named benchmark profile.
- Benchmark
Record - Serializable summary of one algorithm run on one immutable problem.
Constants§
- BENCHMARK_
PROFILES - Built-in benchmark profile metadata in display order.
- BENCHMARK_
RECORD_ FORMAT_ VERSION - Current JSON/CSV benchmark record schema version.
- CPU_
BENCHMARK_ PROFILE - Name of the more representative CPU timing profile.
- SMOKE_
BENCHMARK_ PROFILE - Name of the fast offline synthetic validation profile.
Functions§
- annotate_
benchmark_ profile - Adds the selected profile metadata to one benchmark record.
- benchmark_
profile - Looks up a built-in profile by its exact stable name.
- run_
benchmark_ case - Runs one concrete algorithm and always returns a record. Reconstruction or
metric failures are stored in
record.error; successful reconstruction data is returned separately so callers may inspect or save it. - save_
benchmark_ outputs - Saves the standard reconstruction artifacts for a benchmark case and adds their paths to the record.
- write_
benchmark_ csv - Writes one flattened CSV row per benchmark record.
- write_
benchmark_ json - Writes a versioned pretty-printed JSON report containing
records.