mt op• データ種: signal → table
• 呼び出し: import fullseye as fs; fs.ledger.spc_mt_sn_ratio(md) (実装を直接呼ぶなら import spc; spc.spc_mt_sn_ratio(md)、台帳から引くなら opsspc.get("spc_mt_sn_ratio"))
異常標本の距離から望大特性の SN 比(dB)を出す —— 項目選択の採点。
> 以下の詳細説明は原文のままです —— 要約と見出しは訳出済み。
MT-method item selection asks which features earn their place: run the unit
space with a subset of features, score the *abnormal* samples, and keep the
subset whose distances are largest. The figure of merit is the
larger-the-better signal-to-noise ratio
eta = -10 log10( (1/n) sum_i 1 / MD_i^2 )
an inverse-square average, so one abnormal sample that the subset fails to
separate drags the whole score down — which is the behaviour wanted when the
cost of a miss dominates. The caller drives the subsets (classically by an
orthogonal array); this operator is the closed-form score for one subset.
`md` is a 1-D array of distances for the abnormal samples. Returns a dict
with `sn_ratio_db, the harmonic_mean_md_sq it came from, and n`.
For a constant `MD == d the sum collapses and eta == 20 log10(d)`
exactly — pinned in the tests, and the reason a subset that puts every
abnormal sample at distance 10 scores exactly 20 dB.
Raises `ValueError`: an empty or non-1-D *md*, a non-positive entry (a
zero distance would divide by zero: an abnormal sample sitting exactly on the
unit-space centre means the subset separates nothing), or non-finite input.
• サンプルデータ カタログ(DL URL / ライセンス) — 2-D は skimage.data(BSD/public)+ 合成、3-D は実データ源(Stanford/PDS 等)の DL URL。
• 演算子の来歴・参考文献 — この op 族の元になった研究/手法の出典。
• アルゴリズムの正典(著者・年)と用途は上記ファミリ使い方ガイドに記載。
• poc_spc — py -3.11 examples/poc_spc.py
table を入力に取れる)msa_anova_table · msa_gauge_rr · msa_bias_linearity · msa_attribute_agreement · gum_standard_uncertainty · gum_propagate · gum_expanded · gum_monte_carlo
mt)spc_mt_unit_space · spc_mt_distance
*Provenance: spc.py — SPC operator registry. この per-op ノートは tools/opdocs.py md が自動生成(手編集しない)。*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.