wink-naive-bayes-text-classifier - npm Package Compare versions

Comparing version 1.1.1 to 1.1.2

package.json

 {
   "name": "wink-naive-bayes-text-classifier",
-  "version": "1.1.1",
+  "version": "1.1.2",
   "description": "Configurable Naive Bayes Classifier for text with cross-validation support",
@@ -33,15 +33,15 @@ "keywords": [
   "devDependencies": {
-    "chai": "^3.5.0",
+    "chai": "^4.1.0",
     "coveralls": "^2.11.15",
     "docco": "^0.7.0",
-    "eslint": "^3.13.1",
+    "eslint": "^4.3.0",
     "istanbul": "^0.4.5",
-    "jshint": "^2.9.4",
-    "mocha": "^3.1.0",
+    "jshint": "^2.9.5",
+    "mocha": "^3.4.2",
     "mocha-lcov-reporter": "^1.2.0",
-    "wink-nlp-utils": "^1.0.2"
+    "wink-nlp-utils": "^1.2.1"
   },
   "dependencies": {
-    "wink-helpers": "^1.0.0"
+    "wink-helpers": "^1.1.1"
   }
 }

README.md

@@ -77,3 +77,3 @@
 #### defineConfig( config )
-Defines the configuration from the `config` object. This object must define 2 properties viz. (a) `considerOnlyPresence` and `smoothingFactor`. The `considerOnlyPresence` must be a boolean — true indicates a binarized model; default value is false. The `smoothingFactor` defines the value for additive smoothing; its default value is 0.5. The `defineConfig()` must be called before attempting to learn.
+Defines the configuration from the `config` object. This object must define 2 properties viz. (a) `considerOnlyPresence` and `smoothingFactor`. The `considerOnlyPresence` must be a boolean — true indicates a binarized model; default value is false. The `smoothingFactor` defines the value for additive smoothing; its default value is **1**. The `defineConfig()` must be called before attempting to learn.
 
@@ -104,4 +104,4 @@ #### learn( input, label )
 [
-  [ 'prepay', 12.052329801050746 ],
-  [ 'autoloan', -0.5258305619141872 ]
+  [ 'prepay', 6.169686751688911 ],
+  [ 'autoloan', -6.169686751688911 ]
 ]
@@ -108,0 +108,0 @@ ```

src/wink-naive-bayes-text-classifier.js

@@ -85,3 +85,4 @@ //     wink-naive-bayes-text-classifier
   config.considerOnlyPresence = false;
-  config.smoothingFactor = 0.5;
+  // Default smoothingFactor is set to Laplace add+1 smoothing.
+  config.smoothingFactor = 1;
 
@@ -106,14 +107,15 @@ // ### Private functions
   var logLikelihood = function ( w, label ) {
-    // If there is a **non-zero** `smoothingFactor`, then use the regular
-    // formula for computation. When it is **0**, in that case if the `w`
-    // is not found in vocabulary, return 0; otherwise perform add-1.
-    // Note, a 0 `smoothingFactor` can lead to `unknown` prediction if non-zero
-    // of the words are found in the vocabulary.
+    // To avoid recomputation.
+    var clw = ( count[ label ][ w ] || 0 );
     return (
       ( config.smoothingFactor > 0 ) ?
-        ( Math.log2( ( ( count[ label ][ w ] || 0 ) + config.smoothingFactor ) ) -
-                Math.log2( words[ label ] + ( voc.size * config.smoothingFactor ) ) ) :
-        voc.has( w ) ?  ( Math.log2( ( ( count[ label ][ w ] || 0 ) + 1 ) ) -
-                Math.log2( ( words[ label ] + voc.size ) ) ) :
-                0
+        // Numerator will never be **0** due to smoothing.
+        ( Math.log2( ( clw + config.smoothingFactor ) ) -
+          Math.log2( words[ label ] + ( voc.size * config.smoothingFactor ) ) ) :
+        // Numerator will be 0 if `w` is not found under the `label`.
+        ( clw ) ?
+          // Non-zero numerator means normal handling
+          ( Math.log2( clw ) - Math.log2( ( words[ label ] + voc.size ) ) ) :
+          // Zero numerator: return **0**.
+          0
     );
@@ -124,3 +126,3 @@ }; // logLikelihood()
 
-  // Computes the 1+ smoothed log likelihood `( w | label )`.
+  // Computes the pre-definable smoothed inverse log likelihood `( w | label )`.
   var inverseLogLikelihood = function ( w, label ) {
@@ -141,9 +143,15 @@ // Index and temporary label.
     }
-    // No need to perform `voc.has( w )` check as `odds()` will not call the
-    // `inverseLogLikelihood()` if `logLikelihood()` returns a **0**. It does
-    // so to avoid recomputation. See comments in `logLikelihood()`.
-    return ( Math.log2( ( clw + ( config.smoothingFactor || 1 ) ) ) -
-              Math.log2( ( wl + ( voc.size * ( config.smoothingFactor || 1 ) ) ) )
+
+    return (
+      ( config.smoothingFactor > 0 ) ?
+        // Numerator will never be **0** due to smoothing.
+        ( Math.log2( ( clw + config.smoothingFactor ) ) -
+          Math.log2( wl + ( voc.size * config.smoothingFactor ) ) ) :
+        // Numerator may be 0.
+        ( clw ) ?
+          // Non-zero numerator means normal handling
+          ( Math.log2( clw ) - Math.log2( ( wl + voc.size ) ) ) :
+          // Zero numerator: return **0**.
+          0
     );
-
   }; // inverseLogLikelihood()
@@ -167,2 +175,9 @@
 
+    // Filter unknown tokens.
+    var ivTokens = tokens.filter( function ( e ) {
+      return voc.has( e );
+    } );
+    // No known tokens means simply return **0**.
+    if ( ivTokens.length === 0 ) return 0;
+
     // Compute `samplesNotInLabel`.
@@ -176,6 +191,6 @@ for ( i = 0; i < labelCount; i += 1 ) {
     // Update them for the given tokens for `label`
-    for ( i = 0, imax = tokens.length; i < imax; i += 1 ) {
-      lh += logLikelihood( tokens[ i ], label );
+    for ( i = 0, imax = ivTokens.length; i < imax; i += 1 ) {
+      lh += logLikelihood( ivTokens[ i ], label );
       // If `lh` is **0** then ilh will be zero - avoid computation.
-      ilh += ( lh === 0 ) ? 0 : inverseLogLikelihood( tokens[ i ], label );
+      ilh += ( lh === 0 ) ? 0 : inverseLogLikelihood( ivTokens[ i ], label );
     }
@@ -216,4 +231,11 @@
                                     cfg.considerOnlyPresence : false;
-    config.smoothingFactor = ( isNaN( cfg.smoothingFactor ) ) ?
-            0 : Math.max( Math.min( cfg.smoothingFactor, 1 ), 0 );
+
+    // If smoothing factor is undefined set it to lapalce add+1 smoothing.
+    var sf = ( cfg.smoothingFactor === undefined ) ? 1 : parseFloat( cfg.smoothingFactor );
+    // Throw error for a value beyond 0-1 or NaN.
+    if ( isNaN( sf ) || ( sf < 0 ) || ( sf > 1 ) ) {
+      throw Error( 'winkNBTC: smoothing factor must be a number between 0 & 1, instead found: ' + JSON.stringify( sf ) );
+    }
+    // All good, set smoothingFactor as `sf`.
+    config.smoothingFactor = sf;
     return true;
@@ -220,0 +242,0 @@ }; // defineConfig()

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

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

62060

increased by0.98%

Lines of code

518

increased by4.02%

Dependency changes

• Updatedwink-helpers@^1.1.1