fix: update weights in control points zones

src/utils/__tests__/calculateAdaptiveWeights.test.ts

@@ -0,0 +1,147 @@
+import { expect, test } from 'vitest';
+
+import { calculateAdaptiveWeights } from '../calculateAdaptiveWeights.ts';
+
+test('basic functionality with default options', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1.1, 2.1, 3.1, 4.1, 5.1]);
+  const weights = new Float64Array([1, 1, 1, 1, 1]);
+
+  const result = calculateAdaptiveWeights(yData, baseline, weights, {});
+
+  expect(result).toBeInstanceOf(Float64Array);
+  expect(result).toHaveLength(5);
+  // First and last weights should be 1
+  expect(result[0]).toBe(1);
+  expect(result[4]).toBe(1);
+  // Middle weights should be updated
+  expect(result[1]).toBeLessThan(1);
+  expect(result[2]).toBeLessThan(1);
+  expect(result[3]).toBeLessThan(1);
+});
+
+test('learning rate = 0 returns same weights', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1.1, 2.1, 3.1, 4.1, 5.1]);
+  const weights = new Float64Array([1, 1, 1, 1, 1]);
+
+  const result = calculateAdaptiveWeights(yData, baseline, weights, {
+    learningRate: 0,
+  });
+
+  expect(result).toStrictEqual(weights);
+});
+
+test('high learning rate affects weights more', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1.1, 2.1, 3.1, 4.1, 5.1]);
+  const weights1 = new Float64Array([1, 1, 1, 1, 1]);
+  const weights2 = new Float64Array([1, 1, 1, 1, 1]);
+
+  calculateAdaptiveWeights(yData, baseline, weights1, { learningRate: 0.3 });
+  calculateAdaptiveWeights(yData, baseline, weights2, { learningRate: 0.8 });
+
+  // Higher learning rate should result in more different weights from original
+  for (let i = 1; i < 4; i++) {
+    expect(Math.abs(1 - weights2[i])).toBeGreaterThan(
+      Math.abs(1 - weights1[i]),
+    );
+  }
+});
+
+test('custom factorStd affects weight threshold', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1.1, 2.1, 3.1, 4.1, 5.1]);
+  const weights1 = new Float64Array([1, 1, 1, 1, 1]);
+  const weights2 = new Float64Array([1, 1, 1, 1, 1]);
+
+  calculateAdaptiveWeights(yData, baseline, weights1, { factorStd: 2 });
+  calculateAdaptiveWeights(yData, baseline, weights2, { factorStd: 5 });
+
+  // Different factorStd should produce different results
+  let different = false;
+  for (let i = 1; i < 4; i++) {
+    if (weights1[i] !== weights2[i]) {
+      different = true;
+      break;
+    }
+  }
+
+  expect(different).toBe(true);
+});
+
+test('control points increase weights', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1.1, 2.1, 3.1, 4.1, 5.1]);
+  const weights1 = new Float64Array([1, 1, 1, 1, 1]);
+  const weights2 = new Float64Array([1, 1, 1, 1, 1]);
+
+  calculateAdaptiveWeights(yData, baseline, weights1, {});
+  calculateAdaptiveWeights(yData, baseline, weights2, {
+    controlPoints: new Float64Array([0, 1, 0, 1, 0]),
+  });
+
+  // Weights at control points (index 1 and 3) should be higher
+  expect(weights2[1]).toBeGreaterThan(weights1[1]);
+  expect(weights2[3]).toBeGreaterThan(weights1[3]);
+  // Other weights should remain similar
+  expect(weights2[2]).toBeCloseTo(weights1[2], 3);
+});
+
+test('perfect fit (baseline equals yData)', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1, 2, 3, 4, 5]);
+  const weights = new Float64Array([1, 1, 1, 1, 1]);
+
+  const result = calculateAdaptiveWeights(yData, baseline, weights, {});
+
+  expect(result).toBeInstanceOf(Float64Array);
+  expect(result[0]).toBe(1);
+  expect(result[4]).toBe(1);
+
+  // With perfect fit, weights should be high (close to 1 after normalization)
+  for (let i = 1; i < 4; i++) {
+    expect(result[i]).toBeGreaterThan(0.2);
+  }
+});
+
+test('large residuals reduce weights', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1, 2, 10, 4, 5]); // Large residual at index 2
+  const weights = new Float64Array([1, 1, 1, 1, 1]);
+
+  const result = calculateAdaptiveWeights(yData, baseline, weights, {});
+
+  expect(result[0]).toBe(1);
+  expect(result[4]).toBe(1);
+  // The weight at the large residual point should be reduced
+  expect(result[2]).toBeLessThan(result[1]);
+});
+
+test('modifies input weights array', () => {
+  const yData = new Float64Array([1, 2, 3, 4, 5]);
+  const baseline = new Float64Array([1.1, 2.1, 3.1, 4.1, 5.1]);
+  const weights = new Float64Array([1, 1, 1, 1, 1]);
+
+  const result = calculateAdaptiveWeights(yData, baseline, weights, {});
+
+  // Should return the same array (modified in place)
+  expect(result).toStrictEqual(weights);
+});
+
+test('works with different array types', () => {
+  const yData = [1, 2, 3, 4, 5];
+  const baseline = [1.1, 2.1, 3.1, 4.1, 5.1];
+  const weights = [1, 1, 1, 1, 1];
+
+  const result = calculateAdaptiveWeights(
+    yData as any,
+    baseline as any,
+    weights as any,
+    {},
+  );
+
+  expect(result).toBeDefined();
+  expect(result[0]).toBe(1);
+  expect(result[4]).toBe(1);
+});

src/utils/calculateAdaptiveWeights.ts

@@ -23,11 +23,7 @@ export interface CalculateAdaptiveWeightsOptions extends WeightsAndControlPoints
    * @default 0.5
    */
   learningRate?: number;
-  /**
-   * The minimum allowed weight value to prevent weights from becoming too small.
-   * @default 0.01
-   */
-  minWeight?: number;
+
   /**
    * Factor used to calculate the threshold for determining outliers in the residuals.
    * Higher values mean more tolerance for outliers. The default value is based on noise follow the normal distribution
@@ -55,37 +51,35 @@ export function calculateAdaptiveWeights(
   weights: NumberArray,
   options: CalculateAdaptiveWeightsOptions,
 ) {
-  const {
-    controlPoints,
-    factorStd = 3,
-    learningRate = 0.5,
-    minWeight = 0.01,
-  } = options;
+  const { controlPoints, factorStd = 3, learningRate = 0.5 } = options;
+
+  if (learningRate === 0) {
+    return weights;
+  }
+
   const absResiduals = xAbsolute(xSubtract(yData, baseline));
 
   const medAbsRes = xMedian(absResiduals);
   const mad = 1.4826 * medAbsRes;
-  const threshold = factorStd * mad;
+  const threshold = mad > 0 ? factorStd * mad : 1;
 
   const rawWeights = new Float64Array(absResiduals.length);
   for (let i = 0; i < absResiduals.length; i++) {
     rawWeights[i] = Math.exp(-((absResiduals[i] / threshold) ** 2));
   }
 
-  let maxWeight = Number.MIN_SAFE_INTEGER;
-  const newWeights = Float64Array.from(weights);
   const oneMinusLearningRate = 1 - learningRate;
-  for (let i = 0; i < newWeights.length; i++) {
-    if (controlPoints && controlPoints[i] > 0) continue;
-    const weight = Math.max(
-      minWeight,
-      oneMinusLearningRate * weights[i] + learningRate * rawWeights[i],
-    );
-    newWeights[i] = weight;
-    maxWeight = Math.max(maxWeight, weight);
-  }
-  newWeights[0] = maxWeight;
-  newWeights[weights.length - 1] = maxWeight;
+  for (let i = 0; i < weights.length; i++) {
+    let weight = weights[i];
+    weight = (oneMinusLearningRate * weight + learningRate * rawWeights[i]) / 4;
 
-  return newWeights;
+    if (controlPoints && controlPoints[i] > 0) {
+      weight *= 4;
+    }
+
+    weights[i] = weight;
+  }
+  weights[0] = 1;
+  weights[weights.length - 1] = 1;
+  return weights;
 }

src/x/xWhittakerSmoother.ts

@@ -35,12 +35,6 @@ export interface XWhittakerSmootherOptions extends CalculateAdaptiveWeightsOptio
    * @default 0.5
    */
   learningRate?: number;
-
-  /**
-   * Minimum weight value to avoid division by zero or extremely small weights.
-   * @default 0.01
-   */
-  minWeight?: number;
 }
 
 /**
@@ -59,7 +53,6 @@ export function xWhittakerSmoother(
     tolerance = 1e-6,
     factorStd = 3,
     learningRate = 0.5,
-    minWeight = 0.01,
   } = options;
 
   const size = yData.length;
@@ -89,7 +82,6 @@ export function xWhittakerSmoother(
 
     weights = calculateAdaptiveWeights(yData, newBaseline, weights, {
       controlPoints,
-      minWeight,
       learningRate,
       factorStd,
     });