@iyulab/u-doe 0.6.2 → 0.7.0

package/README.md

@@ -204,11 +204,14 @@ Estimate main effects and interactions for a 2-level factorial design.
     { "name": "A",   "columns": [0],    "estimate": 21.6,  "sum_of_squares": 1870.6, "percent_contribution": 45.2 },
     { "name": "A:C", "columns": [0, 2], "estimate": -18.1, "sum_of_squares": 1314.1, "percent_contribution": 31.7 }
   ],
-  "half_normal": [[18.1, 0.57], [21.6, 1.15]] }
+  "half_normal": [{ "term_index": 1, "abs_effect": 18.1, "quantile": 0.57 },
+                  { "term_index": 0, "abs_effect": 21.6, "quantile": 1.15 }] }
 ```
 
 Interaction names join factor names with `":"` (since 0.5.0; previously bare concatenation `"AC"`). `columns` holds the design-matrix column indices of the term's factors — use it for display formatting (e.g. `"A × C"`) instead of parsing `name`.
 
+Each `half_normal` point carries `term_index` — an index into `effects` — so a point can be labelled directly (e.g. `effects[point.term_index].name`). The points are sorted by `|effect|`, a **different order** from `effects` (model-term order), so pairing them positionally (`effects[i]` ↔ `half_normal[i]`) mislabels every point; always use `term_index`.
+
 #### `fit_rsm(design, responses, factor_names) -> RsmModel`
 
 Fit a second-order Response Surface Model via OLS.
@@ -231,13 +234,17 @@ Compute the steepest ascent path from a fitted RSM model.
 
 Compute Derringer-Suich desirability for multiple responses.
 
-**Input:** `specs`: `[{ "goal": "Maximize"|"Minimize"|"Target", "lower": 0, "target": 100, "upper": 100, "s1": 1, "s2": 1 }]`, `responses`: `Float64Array`
+**Input:** `specs`: `[{ "goal": "Maximize"|"Minimize"|"Target", "lower": 0, "target": 100, "upper": 100, "s1": 1, "s2": 1, "importance": 1 }]`, `responses`: `Float64Array`
+
+`s1`/`s2` are curve-shape exponents (`s = 1` linear, `> 1` convex/stricter, `< 1` concave). `importance` (optional, default `1`) is the Derringer-Suich response weight rᵢ — it is **distinct from** the shape exponents: raise `importance` to make a response count more in the aggregate, not to reshape its curve.
 
 **Output:**
 ```json
 { "individual": [0.8, 0.6], "overall": 0.69 }
 ```
 
+`overall` is the importance-weighted geometric mean D = (∏ dᵢ^rᵢ)^(1/Σrᵢ); with all weights at the default `1` this is the plain geometric mean.
+
 #### `two_level_factorial_power(k, p, n_replicates, effect_size, sigma, alpha) -> f64`
 
 Compute statistical power of a 2^(k-p) factorial design. Returns power in [0, 1].

package/node/u_doe.cjs

@@ -73,10 +73,13 @@ exports.definitive_screening = definitive_screening;
  * Compute Derringer-Suich desirability for multiple responses.
  *
  * `specs`: native array of response specification objects, each:
- *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2 }`
+ *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2, importance? }`
+ *   where `s1`/`s2` are curve-shape exponents and the optional `importance`
+ *   (default 1.0) is the Derringer-Suich weight rᵢ for the overall aggregation.
  * `responses`: flat array of observed response values (one per spec).
  *
- * Returns `{ individual: [f64], overall: f64 }`.
+ * Returns `{ individual: [f64], overall: f64 }` where `overall` is the
+ * importance-weighted geometric mean (∏ dᵢ^rᵢ)^(1/Σrᵢ).
  *
  * # Errors
  * Returns an error string if `specs` has the wrong shape (native JS values,
@@ -138,7 +141,10 @@ exports.doe_anova = doe_anova;
  * `max_order`: maximum interaction order (1 = main effects only, 2 = + 2FI, 3 = + 3FI).
  *
  * Returns `{ effects: [{ name, columns, estimate, sum_of_squares, percent_contribution }],
- *            half_normal: [[abs_effect, quantile]] }`.
+ *            half_normal: [{ term_index, abs_effect, quantile }] }`.
+ * Each `half_normal` point carries `term_index` (an index into `effects`) so the
+ * point can be labelled directly — the points are sorted by `|effect|`, a
+ * different order from `effects`, so positional pairing would mislabel them.
  *
  * # Errors
  * Returns an error string if dimensions do not match or an argument has the

package/node/u_doe.d.cts

@@ -42,10 +42,13 @@ export function definitive_screening(k: number): any;
  * Compute Derringer-Suich desirability for multiple responses.
  *
  * `specs`: native array of response specification objects, each:
- *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2 }`
+ *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2, importance? }`
+ *   where `s1`/`s2` are curve-shape exponents and the optional `importance`
+ *   (default 1.0) is the Derringer-Suich weight rᵢ for the overall aggregation.
  * `responses`: flat array of observed response values (one per spec).
  *
- * Returns `{ individual: [f64], overall: f64 }`.
+ * Returns `{ individual: [f64], overall: f64 }` where `overall` is the
+ * importance-weighted geometric mean (∏ dᵢ^rᵢ)^(1/Σrᵢ).
  *
  * # Errors
  * Returns an error string if `specs` has the wrong shape (native JS values,
@@ -81,7 +84,10 @@ export function doe_anova(design: any, responses: Float64Array, factor_names: an
  * `max_order`: maximum interaction order (1 = main effects only, 2 = + 2FI, 3 = + 3FI).
  *
  * Returns `{ effects: [{ name, columns, estimate, sum_of_squares, percent_contribution }],
- *            half_normal: [[abs_effect, quantile]] }`.
+ *            half_normal: [{ term_index, abs_effect, quantile }] }`.
+ * Each `half_normal` point carries `term_index` (an index into `effects`) so the
+ * point can be labelled directly — the points are sorted by `|effect|`, a
+ * different order from `effects`, so positional pairing would mislabel them.
  *
  * # Errors
  * Returns an error string if dimensions do not match or an argument has the

package/package.json

@@ -5,7 +5,7 @@
     "iyulab"
   ],
   "description": "Design of Experiments (DOE) framework: factorial, Plackett-Burman, CCD, Box-Behnken, Taguchi, effects analysis, RSM, and desirability optimization.",
-  "version": "0.6.2",
+  "version": "0.7.0",
   "license": "MIT",
   "repository": {
     "type": "git",

package/u_doe.d.ts

@@ -42,10 +42,13 @@ export function definitive_screening(k: number): any;
  * Compute Derringer-Suich desirability for multiple responses.
  *
  * `specs`: native array of response specification objects, each:
- *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2 }`
+ *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2, importance? }`
+ *   where `s1`/`s2` are curve-shape exponents and the optional `importance`
+ *   (default 1.0) is the Derringer-Suich weight rᵢ for the overall aggregation.
  * `responses`: flat array of observed response values (one per spec).
  *
- * Returns `{ individual: [f64], overall: f64 }`.
+ * Returns `{ individual: [f64], overall: f64 }` where `overall` is the
+ * importance-weighted geometric mean (∏ dᵢ^rᵢ)^(1/Σrᵢ).
  *
  * # Errors
  * Returns an error string if `specs` has the wrong shape (native JS values,
@@ -81,7 +84,10 @@ export function doe_anova(design: any, responses: Float64Array, factor_names: an
  * `max_order`: maximum interaction order (1 = main effects only, 2 = + 2FI, 3 = + 3FI).
  *
  * Returns `{ effects: [{ name, columns, estimate, sum_of_squares, percent_contribution }],
- *            half_normal: [[abs_effect, quantile]] }`.
+ *            half_normal: [{ term_index, abs_effect, quantile }] }`.
+ * Each `half_normal` point carries `term_index` (an index into `effects`) so the
+ * point can be labelled directly — the points are sorted by `|effect|`, a
+ * different order from `effects`, so positional pairing would mislabel them.
  *
  * # Errors
  * Returns an error string if dimensions do not match or an argument has the

package/u_doe_bg.js

@@ -68,10 +68,13 @@ export function definitive_screening(k) {
  * Compute Derringer-Suich desirability for multiple responses.
  *
  * `specs`: native array of response specification objects, each:
- *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2 }`
+ *   `{ goal: "Maximize"|"Minimize"|"Target", lower, target, upper, s1, s2, importance? }`
+ *   where `s1`/`s2` are curve-shape exponents and the optional `importance`
+ *   (default 1.0) is the Derringer-Suich weight rᵢ for the overall aggregation.
  * `responses`: flat array of observed response values (one per spec).
  *
- * Returns `{ individual: [f64], overall: f64 }`.
+ * Returns `{ individual: [f64], overall: f64 }` where `overall` is the
+ * importance-weighted geometric mean (∏ dᵢ^rᵢ)^(1/Σrᵢ).
  *
  * # Errors
  * Returns an error string if `specs` has the wrong shape (native JS values,
@@ -131,7 +134,10 @@ export function doe_anova(design, responses, factor_names, effect_names) {
  * `max_order`: maximum interaction order (1 = main effects only, 2 = + 2FI, 3 = + 3FI).
  *
  * Returns `{ effects: [{ name, columns, estimate, sum_of_squares, percent_contribution }],
- *            half_normal: [[abs_effect, quantile]] }`.
+ *            half_normal: [{ term_index, abs_effect, quantile }] }`.
+ * Each `half_normal` point carries `term_index` (an index into `effects`) so the
+ * point can be labelled directly — the points are sorted by `|effect|`, a
+ * different order from `effects`, so positional pairing would mislabel them.
  *
  * # Errors
  * Returns an error string if dimensions do not match or an argument has the