sensitivity op• Data kinds: design × signal → table
• Call: import fullseye as fs; fs.ledger.sens_sobol(design=None, y=None, n_boot=500, level=0.95, seed=0) -> 'dict' (to call the implementation directly, import sensitivity; sensitivity.sens_sobol(design=None, y=None, n_boot=500, level=0.95, seed=0) -> 'dict'; from the registry, opsmultivar.get("sens_sobol"))
First-order and total Sobol' indices with bootstrap confidence intervals.
From the outputs on a doe_saltelli design (blocks A, B, A_B^(i)):
• S1_i = (1/N) sum_j f(B)_j (f(A_B^(i))_j - f(A)_j) / V (Saltelli et al. 2010, Table 2 (b));
• ST_i = (1/(2N)) sum_j (f(A)_j - f(A_B^(i))_j)^2 / V (Jansen 1999; Saltelli et al. 2010, Table 2 (f));
• V = variance of the outputs at A and B together.
S1 is the share of output variance a factor explains alone; ST adds every interaction it takes part in. ST ~ 0 means the factor can be fixed anywhere in its range; ST - S1 large means it acts through interactions. Confidence intervals come from n_boot bootstrap resamples of the N rows (the same row index across A, B and A_B^(i)) at level. The sum of S1 (sum_s1) is <= 1 up to estimation error; a value far below 1 says interactions carry the rest.
• design_of_experiments 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.
• sensitivity_tour — py -3.11 examples/sensitivity_tour.py
table as input)—
sensitivity)doe_morris · sens_morris · doe_saltelli
*Provenance: sensitivity.py — MULTIVAR 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.