API

Everything public is importable from pynrpf. The names a pipeline needs:

name	what it is
run(frame, *, columns=None, c=0.7, calibration=RELEASE_CALIBRATION, phi=RELEASE_PHI) -> Result	the whole method on a pandas frame; see Quick start
Result	site_days and intervals frames, provenance, and summary()
Provenance	version, method, calibration_intercept, calibration_slope, calibration_note, phi, c; as_dict() for JSON or a table row
provenance(c=0.7, calibration=RELEASE_CALIBRATION, phi=RELEASE_PHI) -> Provenance	the record run attaches to its result; a Spark job builds it with the arguments it gave run_per_site
pynrpf.spark.run_per_site(sdf, ...)	the same run on a Spark frame; see Spark

Show details

Every other public name

name	what it is
score_siteday(y, s, phi=RELEASE_PHI) -> Score	steps 1 to 7 on one day: 96-slot arrays of net load and solar (MW)
Score	input_ok, n_admissible, winner, runner_up (each a Candidate with start, end, evidence), evidence (r*)
Calibration(intercept, slope, provenance="")	step 8; probability(evidence), evidence_at(p)
RELEASE_CALIBRATION	the Beta-8 pair, intercept −4.590, slope 2.162
RELEASE_PHI	the evidence floor, 8.03e-9 MW
fit_calibration(evidence, labels, slope=None, provenance="")	fit the intercept with the slope fixed, or both numbers (needs scikit-learn)
pynrpf.schemas.SITE_DAYS, INTERVALS	the output tables as (name, dtype, spark type, meaning) tuples

Step modules and errors

The step modules under pynrpf.m9 (stories, windows, edges, bridge, misfit, evidence, winner, calibration, decision) are importable for research and are documented in Method; pynrpf.impact holds the energy and minimum-demand numbers. Errors: KeyError for a missing column, ValueError for timestamps off the grid, duplicated readings, or an invalid c.