torchcell.knowledge_graphs.build_time_projection.calibrate

calibrate(timings: CalibrationTimings, dataset_full_records: dict[str, int], *, calibration_config: SubsetConfig | None = None) → list[AdapterRate]

Fit a per-record rate for each adapter from the calibration build.

records_966 for each adapter is what it processed under the calibration config – min(cap, full) – so dmf/dmi Costanzo use 100000 and every other adapter its full record count. rate = total_sec / records_966.