fit op• Data kinds: simmodel → calresult
• Call: import fullseye as fs; fs.ledger.simcal_fit(model=None, method='lm', n_starts=16, fix=None, threshold=10.0, allow_undetermined=False, seed=0) -> 'dict' (to call the implementation directly, import simcal; simcal.simcal_fit(model=None, method='lm', n_starts=16, fix=None, threshold=10.0, allow_undetermined=False, seed=0) -> 'dict'; from the registry, opssimcal.get("simcal_fit"))
Fit the free knobs of an observed simmodel to its images (weighted least squares) -> calresult.
Minimises chi^2 = sum over valid pixels of (observed − simulated)^2 / variance, the variance coming from the noise model at the current prediction (re-weighted each
iteration: the quasi-likelihood estimating equation, unbiased for a signal-dependent variance). Knobs live in u in [0, 1] (log knobs on a log axis).
Starting point: the initial knobs and n_starts Latin-hypercube points over the ranges are scored by chi^2 and the best is used (the design's "start from the best point of the plan"). method: "lm" (Levenberg-Marquardt, Marquardt-scaled damping; a raw step that leaves the range is not taken — its projection onto the range is tried, a knob on a range end whose descent points outward is held there, and a point the forward model refuses rejects the step and raises the damping, as in lensopt), "nm" (Nelder-Mead; a point outside the range scores the metric's worst value via metriccontract.value_or_worst), "multistart" (LM from the best n_starts // 4 plan points, at least 2, ranked with metriccontract.best_of — for multi-modal problems such as thin-film fringe orders). fix = {knob: value} pins knobs for this fit.
Refusal (metriccontract.MetricContractError, a ValueError) when, at the end point, a knob has no influence (a full-range move changes chi^2 by less than 3.84), is confounded with others (collinearity index gamma_K > threshold, sensitivity.mv_confounding), or sits within 1 % of a range end — the message names the knobs, the reason and the combinations the data do fix. With allow_undetermined=True those knobs get no value (undetermined holds the reason, status = partial) and determined_combos carries the combinations with their values (a product of powers for log knobs). No noise model / no observed images:
refused.
• sim_to_real_calibration family guide
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• poc_pocket_interreflection — py -3.11 examples/poc_pocket_interreflection.py
• poc_simcal_thin_film — py -3.11 examples/poc_simcal_thin_film.py
• simcal_checks_tour — py -3.11 examples/simcal_checks_tour.py
• simcal_tour — py -3.11 examples/simcal_tour.py
calresult as input)simcal_identifiability · simcal_uncertainty · simcal_validate · simcal_discrepancy · simcal_report
fit)*Provenance: simcal.py — SIMCAL operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.