Submission: knn_baseline (RAIL kNN, tasksets 1 & 2)

[marcelo-alvarez]

Submission: knn_baseline

Classic photometric-redshift baseline using RAIL KNearNeighEstimator (k-NN in 9-band ugrizy+YJH magnitude space).

Method

• Per-object p(z) via rail.estimation.algos.k_nearneigh (LSST + Roman bands, catalog tags cardinal_roman_rubin / flagship_roman_rubin).
• Config: zmax=3.0, nzbins=151, trainfrac=0.2, nneigh 3–5, ngrid_sigma=6, NaN non-detections handled via nondetect_val=nan.
• Covers tasksets 1 & 2 (the scoring.py default graded set) × sims {cardinal, flagship} × scenarios {1yr, 10yr} = 8 estimate files + 8 trained models.

Validation

• Each estimate file passes all submit_utils.check_pz_submission_file flags [1–7] locally (valid qp ensemble, zmode + object_id ancil, object_ids match the test file).

Submission tarball (estimates + models) hosted at the release URL referenced in tests/test_knn_baseline.py.

[eacharles]

Thank you.