wink-helpers - npm Package Compare versions

Comparing version 1.4.0 to 1.5.0

package.json

 {
   "name": "wink-helpers",
-  "version": "1.4.0",
-  "description": "Low level helper functions for Javascript array, object and string",
+  "version": "1.5.0",
+  "description": "Functions for cross validation, shuffle, cartesian product and more",
   "keywords": [
@@ -17,3 +17,5 @@ "Array",
     "Shuffle",
-    "Random Shuffle"
+    "Random Shuffle",
+    "Cross Validation",
+    "Machine Learning"
   ],
@@ -41,8 +43,8 @@ "main": "src/wink-helpers.js",
     "chai": "^4.1.2",
-    "coveralls": "^3.0.1",
+    "coveralls": "^3.0.2",
     "docco": "^0.8.0",
-    "eslint": "^4.19.1",
-    "istanbul": "^0.4.5",
+    "eslint": "^5.0.1",
+    "istanbul": "^1.1.0-alpha.1",
     "jshint": "^2.9.5",
-    "mocha": "^5.1.1",
+    "mocha": "^5.2.0",
     "mocha-lcov-reporter": "^1.3.0"
@@ -49,0 +51,0 @@ },

README.md

 
 # wink-helpers
-Helper functions for Javascript array, object, and string
+Functions for cross validation, shuffle, cartesian product and more
 
@@ -9,3 +9,3 @@ ### [![Build Status](https://api.travis-ci.org/winkjs/wink-helpers.svg?branch=master)](https://travis-ci.org/winkjs/wink-helpers) [![Coverage Status](https://coveralls.io/repos/github/winkjs/wink-helpers/badge.svg?branch=master)](https://coveralls.io/github/winkjs/wink-helpers?branch=master) [![Inline docs](http://inch-ci.org/github/winkjs/wink-helpers.svg?branch=master)](http://inch-ci.org/github/winkjs/wink-helpers) [![devDependencies Status](https://david-dm.org/winkjs/wink-helpers/dev-status.svg)](https://david-dm.org/winkjs/wink-helpers?type=dev)
 
-Perform commonly needed operations of array, object and string using **`wink-helpers`**. It is a part of [wink](http://winkjs.org/) — a growing family of high quality packages for Statistical Analysis, Natural Language Processing and Machine Learning in NodeJS.
+Perform cross validation for machine learning, shuffle and many other operations using **`wink-helpers`**. It is a part of [wink](http://winkjs.org/) — a growing family of high quality packages for Statistical Analysis, Natural Language Processing and Machine Learning in NodeJS.
 
@@ -139,3 +139,11 @@
 
+#### cross( classLabels )
+Performs cross validation and generates detailed performance metrics along with the confusion matrix. It is a higher order function that returns an object containing `evaluate()`, `metrics()`,  and `reset()` functions. The `classLabels` should be an array containing all the class labels that may be predicted.
 
+The `evaluate()` function accepts two parameters viz. `truth` — the actual label and `guess` — the predicted label. It is typically called for every row of validation dataset. The evaluation may fail if `truth` or `guess` value is not a valid `classLabels`; or if guess is equal to `unknown`.
+
+The `metrics()` returns an object containing macro-averaged `avgPrecision`, `avgRecall`,  `avgFMeasure` values along with other details such as label-wise recall/precision/f-measure values and the confusion matrix. A value of `null` is returned if no evaluate() has been called before.
+
+The `reset()` re-initializes the current instance for another round of evaluation; the class labels defined at instance creation time are not touched.
+
 ## Need Help?
@@ -142,0 +150,0 @@ If you spot a bug and the same has not yet been reported, raise a new [issue](https://github.com/winkjs/wink-helpers/issues) or consider fixing it and sending a pull request.

src/wink-helpers.js

@@ -294,2 +294,178 @@ //     wink-helpers
 
+// ### cross validation
+/**
+ *
+ * Creates an instance of cross validator useful for machine learning tasks.
+ *
+ * @param {string[]} classLabels - array containing all the class labels.
+ * @return {methods} object conatining set of API methods for tasks like evalutaion,
+ * reset and metrics generation.
+*/
+helpers.validate.cross = function ( classLabels ) {
+  // wink's const for unknown predictions!
+  const unknown = 'unknown';
+  // To ensure that metrics is not computed prior to evaluation.
+  var evaluated = false;
+  // The confusion matrix.
+  var cm;
+  var precision;
+  var recall;
+  var fmeasure;
+
+  // The class labels is assigned to this variable.
+  var labels;
+  // The length of `labels` array.
+  var labelCount;
+  var labelsObj = Object.create( null );
+
+  // Returned!
+  var methods = Object.create( null );
+
+
+  /**
+   *
+   * Resets the current instance for another round of evaluation; the class
+   * labels defined at instance creation time are not touched.
+   *
+   * @return {undefined} nothing!
+  */
+  var reset = function ( ) {
+    evaluated = false;
+    cm = Object.create( null );
+    precision = Object.create( null );
+    recall = Object.create( null );
+    fmeasure = Object.create( null );
+
+    // Initialize confusion matrix and metrics.
+    for ( let i = 0; i < labelCount; i += 1 ) {
+      const row = labels[ i ];
+      labelsObj[ row ] = true;
+      cm[ row ] = Object.create( null );
+      precision[ row ] = 0;
+      recall[ row ] = 0;
+      fmeasure[ row ] = 0;
+      for ( let j = 0; j < labelCount; j += 1 ) {
+        const col = labels[ j ];
+        cm[ row ][ col ] = 0;
+      }
+    }
+  }; // reset()
+
+  /**
+   *
+   * Creates an instance of cross validator useful for machine learning tasks.
+   *
+   * @param {string} truth - the actual class label.
+   * @param {string} guess - the predicted class label.
+   * @return {boolean} returns true if the evaluation is successful. The evaluation
+   * may fail if `truth` or `guess` is not in the array `classLabels` provided at
+   * instance creation time; or if guess is equal to `unknown`.
+  */
+  var evaluate = function ( truth, guess ) {
+    // If prediction failed then return false!
+    if ( guess === unknown || !labelsObj[ truth ] || !labelsObj[ guess ] ) return false;
+    // Update confusion matrix.
+    if ( guess === truth ) {
+      cm[ truth ][ guess ] += 1;
+    } else {
+      cm[ guess ][ truth ] += 1;
+    }
+    evaluated = true;
+    return true;
+  }; // evaluate()
+
+  /**
+   *
+   * It computes a detailed metrics consisting of macro-averaged precision,
+   * recall and f-measure along with their label-wise values and the confusion
+   * matrix.
+   *
+   * @return {object} object containing macro-averaged `avgPrecision`, `avgRecall`,
+   * `avgFMeasure` values along with other details such as label-wise values
+   * and the confusion matrix. A value of `null` is returned if no evaluate()
+   * has been called before.
+  */
+  var metrics = function ( ) {
+    if ( !evaluated ) return null;
+    // Numerators for every label; they are same for precision & recall both.
+    var n = Object.create( null );
+    // Only denominators differs for precision & recall
+    var pd = Object.create( null );
+    var rd = Object.create( null );
+    // `row` and `col` of confusion matrix.
+    var col, row;
+    var i, j;
+    // Macro average values for metrics.
+    var avgPrecision = 0;
+    var avgRecall = 0;
+    var avgFMeasure = 0;
+
+    // Compute label-wise numerators & denominators!
+    for ( i = 0; i < labelCount; i += 1 ) {
+      row = labels[ i ];
+      for ( j = 0; j < labelCount; j += 1 ) {
+        col = labels[ j ];
+        if ( row === col ) {
+          n[ row ] = cm[ row ][ col ];
+        }
+        pd[ row ] = cm[ row ][ col ] + ( pd[ row ] || 0 );
+        rd[ row ] = cm[ col ][ row ] + ( rd[ row ] || 0 );
+      }
+    }
+    // Ready to compute metrics.
+    for ( i = 0; i < labelCount; i += 1 ) {
+      row = labels[ i ];
+      precision[ row ] = +( n[ row ] / pd[ row ] ).toFixed( 4 );
+      // NaN can occur if a label has not been encountered.
+      if ( isNaN( precision[ row ] ) ) precision[ row ] = 0;
+
+      recall[ row ] = +( n[ row ] / rd[ row ] ).toFixed( 4 );
+      if ( isNaN( recall[ row ] ) ) recall[ row ] = 0;
+
+      fmeasure[ row ] = +( 2 * precision[ row ] * recall[ row ] / ( precision[ row ] + recall[ row ] ) ).toFixed( 4 );
+      if ( isNaN( fmeasure[ row ] ) ) fmeasure[ row ] = 0;
+    }
+    // Compute thier averages, note they will be macro avegages.
+    for ( i = 0; i < labelCount; i += 1 ) {
+      avgPrecision += ( precision[ labels[ i ] ] / labelCount );
+      avgRecall += ( recall[ labels[ i ] ] / labelCount );
+      avgFMeasure += ( fmeasure[ labels[ i ] ] / labelCount );
+    }
+    // Return metrics.
+    return (
+      {
+        // Macro-averaged metrics.
+        avgPrecision: +avgPrecision.toFixed( 4 ),
+        avgRecall: +avgRecall.toFixed( 4 ),
+        avgFMeasure: +avgFMeasure.toFixed( 4 ),
+        details: {
+          // Confusion Matrix.
+          confusionMatrix: cm,
+          // Label wise metrics details, from those averages were computed.
+          precision: precision,
+          recall: recall,
+          fmeasure: fmeasure
+        }
+      }
+    );
+  }; // metrics()
+
+  if ( !helpers.validate.isArray( classLabels ) ) {
+    throw Error( 'cross validate: class labels must be an array.' );
+  }
+  if ( classLabels.length < 2 ) {
+    throw Error( 'cross validate: at least 2 class labels are required.' );
+  }
+  labels = classLabels;
+  labelCount = labels.length;
+
+  reset();
+
+  methods.reset = reset;
+  methods.evaluate = evaluate;
+  methods.metrics = metrics;
+
+  return methods;
+}; // cross()
+
 // ### Object Helpers
@@ -296,0 +472,0 @@

New alerts

Long strings

Supply chain risk

Contains long string literals, which may be a sign of obfuscated or packed code.

Found 1 instance in 1 package

URL strings

Supply chain risk

Package contains fragments of external URLs or IP addresses, which the package may be accessing at runtime.

Found 1 instance in 1 package

Improved metrics

Total package byte prevSize

57279

13.49%

Lines of code

438

58.12%

Number of lines in readme file

156

5.41%

Worsened metrics

Number of low supply chain risk alerts

2

Infinity%

No dependency changes