ml-levenberg-marquardt - npm Package Compare versions

Comparing version 1.0.3 to 1.0.4

History.md

@@ -0,1 +1,11 @@
+## [1.0.4](https://github.com/mljs/levenberg-marquardt/compare/v1.0.3...v1.0.4) (2019-05-09)
+
+
+### Bug Fixes
+
+* pluralize min/maxValues parameters & update README ([#20](https://github.com/mljs/levenberg-marquardt/issues/20)) ([afbbd90](https://github.com/mljs/levenberg-marquardt/commit/afbbd90))
+* returns on NaN ([4d61879](https://github.com/mljs/levenberg-marquardt/commit/4d61879))
+
+
+
 <a name="1.0.3"></a>
@@ -2,0 +12,0 @@ ## [1.0.3](https://github.com/mljs/levenberg-marquardt/compare/v1.0.2...v1.0.3) (2018-02-07)

lib/index.js

@@ -76,6 +76,6 @@ 'use strict';
   for (var point = 0; point < m; point++) {
-    ans[point] = data.y[point] - evaluatedData[point];
+    ans[point] = [data.y[point] - evaluatedData[point]];
   }
 
-  return new mlMatrix.Matrix([ans]);
+  return new mlMatrix.Matrix(ans);
 }
@@ -100,12 +100,8 @@
 ) {
-  var identity = mlMatrix.Matrix.eye(params.length).mul(
-    damping * gradientDifference * gradientDifference
-  );
+  var value = damping * gradientDifference * gradientDifference;
+  var identity = mlMatrix.Matrix.eye(params.length, params.length, value);
 
-  var l = data.x.length;
-  var evaluatedData = new Array(l);
   const func = parameterizedFunction(params);
-  for (var i = 0; i < l; i++) {
-    evaluatedData[i] = func(data.x[i]);
-  }
+  var evaluatedData = data.x.map((e) => func(e));
+
   var gradientFunc = gradientFunction(
@@ -118,6 +114,7 @@ data,
   );
-  var matrixFunc = matrixFunction(data, evaluatedData).transposeView();
+  var matrixFunc = matrixFunction(data, evaluatedData);
   var inverseMatrix = mlMatrix.inverse(
-    identity.add(gradientFunc.mmul(gradientFunc.transposeView()))
+    identity.add(gradientFunc.mmul(gradientFunc.transpose()))
   );
+
   params = new mlMatrix.Matrix([params]);
@@ -129,3 +126,3 @@ params = params.sub(
       .mul(gradientDifference)
-      .transposeView()
+      .transpose()
   );
@@ -143,2 +140,4 @@
  * @param {number} [options.gradientDifference = 10e-2] - Adjustment for decrease the damping parameter
+ * @param {Array<number>} [options.minValues] - Minimum allowed values for parameters
+ * @param {Array<number>} [options.maxValues] - Maximum allowed values for parameters
  * @param {Array<number>} [options.initialValues] - Array of initial parameter values
@@ -159,2 +158,4 @@ * @param {number} [options.maxIterations = 100] - Maximum of allowed iterations
     errorTolerance = 10e-3,
+    minValues,
+    maxValues,
     initialValues
@@ -176,7 +177,4 @@ } = options;
     );
-  } else {
-    let dataLen = data.x.length;
-    if (dataLen !== data.y.length) {
-      throw new Error('The data parameter elements must have the same size');
-    }
+  } else if (data.x.length !== data.y.length) {
+    throw new Error('The data parameter elements must have the same size');
   }
@@ -186,3 +184,10 @@
     initialValues || new Array(parameterizedFunction.length).fill(1);
+  let parLen = parameters.length;
+  maxValues = maxValues || new Array(parLen).fill(Number.MAX_SAFE_INTEGER);
+  minValues = minValues || new Array(parLen).fill(Number.MIN_SAFE_INTEGER);
 
+  if (maxValues.length !== minValues.length) {
+    throw new Error('minValues and maxValues must be the same size');
+  }
+
   if (!Array.isArray(parameters)) {
@@ -208,3 +213,12 @@ throw new Error('initialValues must be an array');
     );
+
+    for (let k = 0; k < parLen; k++) {
+      parameters[k] = Math.min(
+        Math.max(minValues[k], parameters[k]),
+        maxValues[k]
+      );
+    }
+
     error = errorCalculation(data, parameters, parameterizedFunction);
+    if (isNaN(error)) break;
     converged = error <= errorTolerance;
@@ -211,0 +225,0 @@ }

package.json

 {
   "name": "ml-levenberg-marquardt",
-  "version": "1.0.3",
+  "version": "1.0.4",
   "description": "Curve fitting method in javascript",
   "main": "./lib/index.js",
-  "files": ["runkit.js", "src", "lib"],
+  "module": "./src/index.js",
+  "files": [
+    "runkit.js",
+    "src",
+    "lib"
+  ],
   "scripts": {
     "eslint": "eslint src",
     "eslint-fix": "npm run eslint -- --fix",
-    "prepublish": "rollup -c",
-    "test": "npm run test-simple && npm run eslint",
-    "test-travis": "jest --coverage && codecov",
-    "test-simple": "jest"
+    "prepare": "rollup -c",
+    "test": "npm run test-only && npm run eslint",
+    "test-coverage": "npm run test-only -- --coverage",
+    "test-only": "jest",
+    "test-travis": "npm run test-coverage && npm run eslint"
   },
@@ -19,3 +25,11 @@ "repository": {
   },
-  "keywords": ["machine", "learning", "data", "mining", "datamining"],
+  "keywords": [
+    "machine",
+    "learning",
+    "data",
+    "mining",
+    "datamining",
+    "levenberg",
+    "marquardt"
+  ],
   "author": "Miguel Asencio <[e-mail]>",
@@ -29,16 +43,20 @@ "license": "MIT",
   "devDependencies": {
-    "babel-plugin-transform-es2015-modules-commonjs": "^6.26.0",
-    "codecov": "^2.3.1",
-    "eslint": "^4.17.0",
-    "eslint-config-cheminfo": "^1.15.1",
-    "eslint-plugin-import": "^2.8.0",
-    "eslint-plugin-jest": "^21.7.0",
-    "jest": "^22.1.4",
-    "jest-matcher-deep-close-to": "^1.1.0",
-    "mocha-better-spec-reporter": "^3.1.0",
-    "rollup": "^0.55.3"
+    "babel-plugin-transform-es2015-modules-commonjs": "^6.26.2",
+    "cz-conventional-changelog": "^2.1.0",
+    "eslint": "^5.16.0",
+    "eslint-config-cheminfo": "^1.20.1",
+    "eslint-plugin-import": "^2.17.1",
+    "eslint-plugin-jest": "^22.4.1",
+    "jest": "^24.7.1",
+    "jest-matcher-deep-close-to": "^1.3.0",
+    "rollup": "^1.10.0"
   },
   "dependencies": {
-    "ml-matrix": "^5.0.1"
+    "ml-matrix": "^5.3.0"
+  },
+  "config": {
+    "commitizen": {
+      "path": "./node_modules/cz-conventional-changelog"
+    }
   }
 }

README.md

@@ -49,5 +49,15 @@ # levenberg-marquardt
 
+// Optionally, restrict parameters to minimum & maximum values
+let minValues = [
+  /* a_min, b_min, c_min, ... */    
+];
+let maxValues = [
+  /* a_max, b_max, c_max, ... */    
+];
+
 const options = {
   damping: 1.5,
   initialValues: initialValues,
+  minValues: minValues,
+  maxValues: maxValues,
   gradientDifference: 10e-2,
@@ -54,0 +64,0 @@ maxIterations: 100,

src/__tests__/errorCalculation.js

 import errorCalculation from '../errorCalculation';
 
-function sinFunction([a, b]) {
-  return (t) => a * Math.sin(b * t);
-}
+describe('parameterError', () => {
+  describe('Linear functions', () => {
+    const linearFunction = ([slope, intercept]) => ((x) => slope * x + intercept);
 
-describe('errorCalculation test', () => {
-  it('Simple case', () => {
-    const len = 20;
-    let data = {
-      x: new Array(len),
-      y: new Array(len)
+    const sampleParameters = [1, 1];
+    const n = 10;
+    const xs = new Array(n).fill(0).map((zero, i) => i);
+    const data = {
+      x: xs,
+      y: xs.map(linearFunction(sampleParameters))
     };
-    let sampleFunction = sinFunction([2, 2]);
-    for (let i = 0; i < len; i++) {
-      data.x[i] = i;
-      data.y[i] = sampleFunction(i);
-    }
 
-    expect(errorCalculation(data, [2, 2], sinFunction)).toBeCloseTo(0, 3);
-    expect(errorCalculation(data, [4, 4], sinFunction)).toBeCloseTo(48.7, 1);
+    it('parameterError should be zero for an exact fit', () => {
+      expect(errorCalculation(data, sampleParameters, linearFunction)).toBeCloseTo(0, 3);
+    });
+
+    it('parameterError should match the sum of absolute difference between the model and the data', () => {
+      const parameters = Array.from(sampleParameters);
+      // Offset line so that it's still parallel but differs by 1 at each point
+      // Then each point will result in a residual increase of 1
+      parameters[1] += 1;
+      expect(errorCalculation(data, parameters, linearFunction)).toBeCloseTo(n, 3);
+    });
   });
 });

src/index.js

@@ -11,2 +11,4 @@ import errorCalculation from './errorCalculation';
  * @param {number} [options.gradientDifference = 10e-2] - Adjustment for decrease the damping parameter
+ * @param {Array<number>} [options.minValues] - Minimum allowed values for parameters
+ * @param {Array<number>} [options.maxValues] - Maximum allowed values for parameters
  * @param {Array<number>} [options.initialValues] - Array of initial parameter values
@@ -27,2 +29,4 @@ * @param {number} [options.maxIterations = 100] - Maximum of allowed iterations
     errorTolerance = 10e-3,
+    minValues,
+    maxValues,
     initialValues
@@ -44,7 +48,4 @@ } = options;
     );
-  } else {
-    let dataLen = data.x.length;
-    if (dataLen !== data.y.length) {
-      throw new Error('The data parameter elements must have the same size');
-    }
+  } else if (data.x.length !== data.y.length) {
+    throw new Error('The data parameter elements must have the same size');
   }
@@ -54,3 +55,10 @@
     initialValues || new Array(parameterizedFunction.length).fill(1);
+  let parLen = parameters.length;
+  maxValues = maxValues || new Array(parLen).fill(Number.MAX_SAFE_INTEGER);
+  minValues = minValues || new Array(parLen).fill(Number.MIN_SAFE_INTEGER);
 
+  if (maxValues.length !== minValues.length) {
+    throw new Error('minValues and maxValues must be the same size');
+  }
+
   if (!Array.isArray(parameters)) {
@@ -76,3 +84,12 @@ throw new Error('initialValues must be an array');
     );
+
+    for (let k = 0; k < parLen; k++) {
+      parameters[k] = Math.min(
+        Math.max(minValues[k], parameters[k]),
+        maxValues[k]
+      );
+    }
+
     error = errorCalculation(data, parameters, parameterizedFunction);
+    if (isNaN(error)) break;
     converged = error <= errorTolerance;
@@ -79,0 +96,0 @@ }

src/step.js

@@ -51,6 +51,6 @@ import { inverse, Matrix } from 'ml-matrix';
   for (var point = 0; point < m; point++) {
-    ans[point] = data.y[point] - evaluatedData[point];
+    ans[point] = [data.y[point] - evaluatedData[point]];
   }
 
-  return new Matrix([ans]);
+  return new Matrix(ans);
 }
@@ -75,12 +75,8 @@
 ) {
-  var identity = Matrix.eye(params.length).mul(
-    damping * gradientDifference * gradientDifference
-  );
+  var value = damping * gradientDifference * gradientDifference;
+  var identity = Matrix.eye(params.length, params.length, value);
 
-  var l = data.x.length;
-  var evaluatedData = new Array(l);
   const func = parameterizedFunction(params);
-  for (var i = 0; i < l; i++) {
-    evaluatedData[i] = func(data.x[i]);
-  }
+  var evaluatedData = data.x.map((e) => func(e));
+
   var gradientFunc = gradientFunction(
@@ -93,6 +89,7 @@ data,
   );
-  var matrixFunc = matrixFunction(data, evaluatedData).transposeView();
+  var matrixFunc = matrixFunction(data, evaluatedData);
   var inverseMatrix = inverse(
-    identity.add(gradientFunc.mmul(gradientFunc.transposeView()))
+    identity.add(gradientFunc.mmul(gradientFunc.transpose()))
   );
+
   params = new Matrix([params]);
@@ -104,3 +101,3 @@ params = params.sub(
       .mul(gradientDifference)
-      .transposeView()
+      .transpose()
   );
@@ -107,0 +104,0 @@

New alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

New author

Supply chain risk

A new npm collaborator published a version of the package for the first time. New collaborators are usually benign additions to a project, but do indicate a change to the security surface area of a package.

Found 1 instance in 1 package

Fixed alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

Improved metrics

Total package byte prevSize

29443

increased by29.49%

Dev dependency count

9

decreased by-10%

Number of package files

12

increased by9.09%

Lines of code

638

increased by20.15%

Number of lines in readme file

86

increased by13.16%

Worsened metrics

Number of medium supply chain risk alerts

1

increased byInfinity%

Dependency changes

• Updatedml-matrix@^5.3.0