js-svm - npm Package Compare versions

Comparing version 1.0.1 to 1.0.2

build/jssvm.min.js

@@ -1,1 +0,1 @@
-var jssvm=jssvm||{};!function(t){t.small=function(){for(var t=.5,r=1*t+1;r>1;)r=1*(t/=2)+1;return 2*t*2},t.quickSort=function(r,s){t._quickSort(r,0,r.length-1,s)},t._quickSort=function(r,s,i,a){if(!(s>=i)){var h=t._partition(r,s,i,a);t._quickSort(r,s,h-1,a),t._quickSort(r,h+1,i,a)}},t._partition=function(r,s,i,a){for(var h=r[s],o=s,e=i+1;;){for(;a(h,r[++o])>0&&!(o>=i););for(;a(h,r[--e])<0&&!(e<=s););if(o>=e)break;t._exchange(r,o,e)}return t._exchange(r,s,e),e},t._exchange=function(t,r,s){var i=t[r];t[r]=t[s],t[s]=i},t.deltaStep=function(t){var r=Math.sqrt(this.small()),s=Math.max(1,Math.abs(t)),i=Math.max(0,Math.abs(r*s));return t<0&&(i=-i),i},t.dotProduct=function(t,r){for(var s=0,i=r.length,a=0;a<i;++a)s+=t[a]*r[a];return s};var r=function(t){(t=t||{}).iterations||(t.iterations=100),t.alpha||(t.alpha=1e-4),t.C||(t.C=1),t.trace||(t.trace=!1),this.iterations=t.iterations,this.alpha=t.alpha,this.C=t.C,this.trace=t.trace};r.prototype.fit=function(t){this.dim=t[0].length;for(var r=t.length,s=[],i=[],a=0;a<r;++a){var h=t[a],o=[],e=h[h.length-1];o.push(1);for(var n=0;n<h.length-1;++n)o.push(h[n]);s.push(o),i.push(e)}this.theta=[];for(c=0;c<this.dim;++c)this.theta.push(0);for(var f=0;f<this.iterations;++f){for(var l=this.grad(s,i,this.theta),c=0;c<this.dim;++c)this.theta[c]=this.theta[c]-this.alpha*l[c];this.trace&&console.log("cost: "+this.cost(s,i,this.theta))}var u=this.transform(t);this.threshold=null;for(var r=t.length,a=0;a<r;++a)1==t[a][this.dim-1]&&(null==this.threshold||this.threshold>u[a])&&(this.threshold=u[a]);return{theta:this.theta,cost:this.cost(s,i,this.theta),threshold:this.threshold,config:{alpha:this.alpha,C:this.C,iterations:this.iterations,trace:this.trace}}},r.prototype.smallStep=function(r,s){return t.deltaStep(r[s])},r.prototype.grad=function(t,r,s){for(var i=[],a=this.cost(t,r,s),h=0;h<this.dim;++h){var o=this.smallStep(s,h),e=s[h];s[h]+=o;var n=this.cost(t,r,s);s[h]=e,i.push((n-a)/2)}return i},r.prototype.cost=function(r,s,i){for(var a=r.length,h=0,o=0;o<a;++o){var e=r[o],n=s[o];h+=this.C*(n*this.cost1(e,n,i)+(1-n)*this.cost0(e,n,i))}return h+=t.dotProduct(i,i)/2,h/=a},r.prototype.cost1=function(r,s,i){var a=t.dotProduct(r,i);return a>1?0:1-a},r.prototype.cost0=function(r,s,i){var a=t.dotProduct(r,i);return a<-1?0:1+a},r.prototype.transform=function(r){if(r[0].length){for(var s=[],i=0;i<r.length;++i){var a=this.transform(r[i]);s.push(a)}return s}var h=[];h.push(1);for(var o=0;o<r.length;++o)h.push(r[o]);return t.dotProduct(h,this.theta)},t.LinearSvm=r;var s=function(r){this.classifier=new t.LinearSvm(r)};s.prototype.fit=function(t){return this.classifier.fit(t)},s.prototype.transform=function(t){return this.classifier.transform(t)>this.threshold?1:0},t.BinarySvmClassifier=s;var i=function(t){(t=t||{}).alpha||(t.alpha=.001),t.iterations||(t.iterations=100),t.C||(t.C=C),this.alpha=t.alpha,this.C=t.C,this.iterations=t.iterations};i.prototype.fit=function(r,s){this.dim=r[0].length;var i=r.length;if(!s){s=[];for(c=0;c<i;++c){for(var a=!1,h=r[c][this.dim-1],o=0;o<s.length;++o)if(h==s[o]){a=!0;break}a||s.push(h)}}this.classes=s,this.classifiers={};for(var e={},n=0;n<this.classes.length;++n){var f=this.classes[n];this.classifiers[f]=new t.LinearSvm({alpha:this.alpha,C:this.C,iterations:this.iterations});for(var l=[],c=0;c<i;++c){for(var u=[],o=0;o<this.dim-1;++o)u.push(r[c][o]);u.push(r[c][this.dim-1]==f?1:0),l.push(u)}e[f]=this.classifiers[f].fit(l)}return e},i.prototype.transform=function(t){if(t[0].length){for(var r=[],s=0;s<t.length;++s){var i=this.transform(t[s]);r.push(i)}return r}for(var a=0,h="",o=0;o<this.classes.length;++o){var e=this.classes[o],n=this.classifiers[e].transform(t)-this.classifiers[e].threshold;a<n&&(a=n,h=e)}return h},t.MultiClassSvmClassifier=i}(jssvm),module&&(module.exports=jssvm);
+var jssvm=jssvm||{};!function(t){t.small=function(){for(var t=.5,s=1*t+1;s>1;)s=1*(t/=2)+1;return 2*t*2},t.quickSort=function(s,i){t._quickSort(s,0,s.length-1,i)},t._quickSort=function(s,i,r,a){if(!(i>=r)){var h=t._partition(s,i,r,a);t._quickSort(s,i,h-1,a),t._quickSort(s,h+1,r,a)}},t._partition=function(s,i,r,a){for(var h=s[i],e=i,n=r+1;;){for(;a(h,s[++e])>0&&!(e>=r););for(;a(h,s[--n])<0&&!(n<=i););if(e>=n)break;t._exchange(s,e,n)}return t._exchange(s,i,n),n},t._exchange=function(t,s,i){var r=t[s];t[s]=t[i],t[i]=r},t.deltaStep=function(t){var s=Math.sqrt(this.small()),i=Math.max(1,Math.abs(t)),r=Math.max(0,Math.abs(s*i));return t<0&&(r=-r),r},t.dotProduct=function(t,s){for(var i=0,r=s.length,a=0;a<r;++a)i+=t[a]*s[a];return i},t.gaussian=function(s,i,r,a){for(var h=i.length,e=0,n=0;n<h;++n){var o=r[n];e+=i[n]*t._gaussian(s,o,a,h)}return e},t._gaussian=function(s,i,r,a){for(var h=[],e=0;e<a;++e)h.push(s[e]-i[e]);var n=t.dotProduct(h,h);return Math.exp(-n/(r*r))};var s=function(t){(t=t||{}).iterations||(t.iterations=1e3),t.alpha||(t.alpha=.01),t.C||(t.C=5),t.trace||(t.trace=!1),this.iterations=t.iterations,this.alpha=t.alpha,this.C=t.C,this.trace=t.trace};s.prototype.fit=function(t){this.dim=t[0].length;for(var s=t.length,i=[],r=[],a=0;a<s;++a){var h=t[a],e=[],n=h[h.length-1];e.push(1);for(var o=0;o<h.length-1;++o)e.push(h[o]);i.push(e),r.push(n)}this.theta=[];for(f=0;f<this.dim;++f)this.theta.push(0);for(var l=0;l<this.iterations;++l){for(var u=this.grad(i,r,this.theta),f=0;f<this.dim;++f)this.theta[f]=this.theta[f]-this.alpha*u[f];this.trace&&console.log("cost: "+this.cost(i,r,this.theta))}var c=this.transform(t);this.threshold=null;for(var s=t.length,a=0;a<s;++a)1==t[a][this.dim-1]&&(null==this.threshold||this.threshold>c[a])&&(this.threshold=c[a]);return{theta:this.theta,cost:this.cost(i,r,this.theta),threshold:this.threshold,config:{alpha:this.alpha,C:this.C,iterations:this.iterations,trace:this.trace}}},s.prototype.smallStep=function(s,i){return t.deltaStep(s[i])},s.prototype.grad=function(t,s,i){for(var r=[],a=this.cost(t,s,i),h=0;h<this.dim;++h){var e=this.smallStep(i,h),n=i[h];i[h]+=e;var o=this.cost(t,s,i);i[h]=n,r.push((o-a)/2)}return r},s.prototype.cost=function(s,i,r){for(var a=s.length,h=0,e=0;e<a;++e){var n=s[e],o=i[e];h+=this.C*(o*this.cost1(n,o,r)+(1-o)*this.cost0(n,o,r))}return h+=t.dotProduct(r,r)/2,h/=a},s.prototype.cost1=function(s,i,r){var a=t.dotProduct(s,r);return a>1?0:1-a},s.prototype.cost0=function(s,i,r){var a=t.dotProduct(s,r);return a<-1?0:1+a},s.prototype.transform=function(s){if(s[0].length){for(var i=[],r=0;r<s.length;++r){var a=this.transform(s[r]);i.push(a)}return i}var h=[];h.push(1);for(var e=0;e<s.length;++e)h.push(s[e]);return t.dotProduct(h,this.theta)},t.LinearSvm=s;var i=function(t){(t=t||{}).iterations||(t.iterations=1e3),t.alpha||(t.alpha=.01),t.C||(t.C=5),t.trace||(t.trace=!1),t.sigma||(t.sigma=1),this.iterations=t.iterations,this.alpha=t.alpha,this.C=t.C,this.trace=t.trace,this.sigma=t.sigma};i.prototype.fit=function(t){this.dim=t[0].length;for(var s=t.length,i=[],r=[],a=0;a<s;++a){var h=t[a],e=[],n=h[h.length-1];e.push(1);for(var o=0;o<h.length-1;++o)e.push(h[o]);i.push(e),r.push(n)}this.theta=[],this.L=[];for(f=0;f<this.dim;++f)this.theta.push(0),this.L.push(i[f]);for(var l=0;l<this.iterations;++l){for(var u=this.grad(i,r,this.theta),f=0;f<this.dim;++f)this.theta[f]=this.theta[f]-this.alpha*u[f];this.trace&&console.log("cost: "+this.cost(i,r,this.theta))}var c=this.transform(t);this.threshold=null;for(var s=t.length,a=0;a<s;++a)1==t[a][this.dim-1]&&(null==this.threshold||this.threshold>c[a])&&(this.threshold=c[a]);return{theta:this.theta,L:this.L,cost:this.cost(i,r,this.theta),threshold:this.threshold,config:{alpha:this.alpha,C:this.C,iterations:this.iterations,trace:this.trace,sigma:this.sigma}}},i.prototype.smallStep=function(s,i){return t.deltaStep(s[i])},i.prototype.grad=function(t,s,i){for(var r=[],a=this.cost(t,s,i),h=0;h<this.dim;++h){var e=this.smallStep(i,h),n=i[h];i[h]+=e;var o=this.cost(t,s,i);i[h]=n,r.push((o-a)/2)}return r},i.prototype.cost=function(s,i,r){for(var a=s.length,h=0,e=0;e<a;++e){var n=s[e],o=i[e];h+=this.C*(o*this.cost1(n,o,r)+(1-o)*this.cost0(n,o,r))}return h+=t.dotProduct(r,r)/2,h/=a},i.prototype.cost1=function(s,i,r){var a=t.gaussian(s,r,this.L,this.sigma);return a>1?0:1-a},i.prototype.cost0=function(s,i,r){var a=t.gaussian(s,r,this.L,this.sigma);return a<-1?0:1+a},i.prototype.transform=function(s){if(s[0].length){for(var i=[],r=0;r<s.length;++r){var a=this.transform(s[r]);i.push(a)}return i}var h=[];h.push(1);for(var e=1;e<this.dim;++e)h.push(s[e-1]);return t.gaussian(h,this.theta,this.L,this.sigma)},t.KernelSvm=i;var r=function(s){(s=s||{}).kernel||(s.kernel="linear"),s.sigma||(s.sigma=1),this.kernel=s.kernel,"linear"==s.kernel?this.classifier=new t.LinearSvm(s):"gaussian"==s.kernel?this.classifier=new t.KernelSvm(s):this.classifier=new t.LinearSvm(s)};r.prototype.fit=function(t){var s=this.classifier.fit(t);return s.kernel=this.kernel,s},r.prototype.transform=function(t){return this.classifier.transform(t)>this.classifier.threshold?1:0},t.BinarySvmClassifier=r;var a=function(t){(t=t||{}).alpha||(t.alpha=.01),t.iterations||(t.iterations=1e3),t.C||(t.C=5),t.kernel||(t.kernel="linear"),t.sigma||(t.sigma=1),this.alpha=t.alpha,this.C=t.C,this.iterations=t.iterations,this.sigma=t.sigma,this.kernel=t.kernel};a.prototype.fit=function(s,i){this.dim=s[0].length;var r=s.length;if(!i){i=[];for(f=0;f<r;++f){for(var a=!1,h=s[f][this.dim-1],e=0;e<i.length;++e)if(h==i[e]){a=!0;break}a||i.push(h)}}this.classes=i,this.classifiers={};for(var n={},o=0;o<this.classes.length;++o){var l=this.classes[o];"linear"==this.kernel?this.classifiers[l]=new t.LinearSvm({alpha:this.alpha,C:this.C,iterations:this.iterations}):"gaussian"==this.kernel?this.classifiers[l]=new t.KernelSvm({alpha:this.alpha,C:this.C,iterations:this.iterations,sigma:this.sigma}):this.classifiers[l]=new t.LinearSvm({alpha:this.alpha,C:this.C,iterations:this.iterations});for(var u=[],f=0;f<r;++f){for(var c=[],e=0;e<this.dim-1;++e)c.push(s[f][e]);c.push(s[f][this.dim-1]==l?1:0),u.push(c)}n[l]=this.classifiers[l].fit(u)}return n.kernel=this.kernel,n},a.prototype.transform=function(t){if(t[0].length){for(var s=[],i=0;i<t.length;++i){var r=this.transform(t[i]);s.push(r)}return s}for(var a=0,h="",e=0;e<this.classes.length;++e){var n,o=this.classes[e];a<=(n=this.classifiers[o].transform(t)-this.classifiers[o].threshold)&&(a=n,h=o)}return h},t.MultiClassSvmClassifier=a}(jssvm),module&&(module.exports=jssvm);

package.json

 {
   "name": "js-svm",
-  "version": "1.0.1",
+  "version": "1.0.2",
   "description": "Package implements linear svm and kernel svm that supports binary and mult-class classification",
@@ -5,0 +5,0 @@ "author": "Xianshun Chen",

README.md

@@ -25,8 +25,3 @@ # js-svm
 
-var svm = new jssvm.BinarySvmClassifier({
-   alpha: 0.01,
-   iterations: 1000,
-   C: 5.0,
-   trace: false
-});
+var svm = new jssvm.BinarySvmClassifier();
 
@@ -63,2 +58,13 @@ iris.shuffle();
 
+To configure the BinarySvmClassifier, use the following code when it is created:
+
+```javascript
+var svm = new jssvm.BinarySvmClassifier({
+   alpha: 0.01, // learning rate
+   iterations: 1000, // maximum iterations
+   C: 5.0, // panelty term
+   trace: false // debug tracing
+});
+```
+
 ### Multi-Class Classification using One-vs-All Logistic Regression
@@ -72,7 +78,3 @@
 
-var classifier = new jssvm.MultiClassSvmClassifier({
-   alpha: 0.001,
-   iterations: 1000,
-   C: 5.0
-});
+var classifier = new jssvm.MultiClassSvmClassifier();
 
@@ -109,3 +111,39 @@ iris.shuffle();
 
+To configure the MultiClassSvmClassifier, use the following code when it is created:
 
+```javascript
+var classifier = new jssvm.MultiClassSvmClassifier({
+   alpha: 0.01, // learning rate
+   iterations: 1000, // maximum iterations
+   C: 5.0 // panelty term
+   sigma: 1.0 // the standard deviation for the gaussian kernel
+});
+```
+
+### Switch between linear and guassian kernel
+
+By default the kernel used by the binary and multi-class classifier is "linear" which can be printed by:
+
+```javascript
+console.log(classifier.kernel);
+```
+
+To switch to use gaussian kernel, put the property 'kernel: "gaussian"' in the config data when the classifier is created:
+
+```javascript
+var svm = new jssvm.BinarySvmClassifier({
+   ...,
+   kernel: 'gaussian'
+});
+
+....
+
+var svm = new jssvm.MultiClassSvmClassifier({
+   ...,
+   kernel: 'gaussian'
+});
+
+```
+
+
 ### Usage In HTML
@@ -112,0 +150,0 @@

src/jssvm.js

@@ -73,2 +73,23 @@ var jssvm = jssvm || {};
     
+    jsr.gaussian = function(x_i, theta, L, sigma) {
+        
+        var dim = theta.length;
+        
+        var sum = 0;
+        for(var d = 0; d < dim; ++d){
+            var l_d = L[d];
+            sum += theta[d] * jsr._gaussian(x_i, l_d, sigma, dim);
+        }
+        return sum;
+    };
+    
+    jsr._gaussian = function(x_i, l_d, sigma, dim){
+        var dx = [];
+        for(var d = 0; d < dim; ++d){
+            dx.push(x_i[d] - l_d[d]);
+        }
+        var z = jsr.dotProduct(dx, dx);
+        return Math.exp(-z / (sigma * sigma));
+    }
+    
 	var LinearSvm = function(config) {
@@ -78,9 +99,9 @@ config = config || {};
         if(!config.iterations){
-            config.iterations = 100;
+            config.iterations = 1000;
         }
         if(!config.alpha){
-            config.alpha = 0.0001;
+            config.alpha = 0.01;
         }
         if(!config.C){
-            config.C = 1.0;
+            config.C = 5.0;
         }
@@ -233,4 +254,185 @@ if(!config.trace){
     
+    var KernelSvm = function(config) {
+        config = config || {};
+        
+        if(!config.iterations){
+            config.iterations = 1000;
+        }
+        if(!config.alpha){
+            config.alpha = 0.01;
+        }
+        if(!config.C){
+            config.C = 5.0;
+        }
+        if(!config.trace){
+            config.trace = false;
+        }
+        if(!config.sigma) {
+            config.sigma = 1.0;
+        }
+        
+        this.iterations = config.iterations;
+        this.alpha = config.alpha;
+        this.C = config.C;
+        this.trace = config.trace;
+        this.sigma = config.sigma;
+    };
+    
+    KernelSvm.prototype.fit = function(data) {
+        this.dim = data[0].length;
+        var N = data.length;
+        
+        var X = [];
+        var Y = [];
+        for(var i=0; i < N; ++i){
+            var row = data[i];
+            var x_i = [];
+            var y_i = row[row.length-1];
+            x_i.push(1.0);
+            for(var j=0; j < row.length-1; ++j){
+                x_i.push(row[j]);
+            }
+            X.push(x_i);
+            Y.push(y_i);
+        }
+        
+        this.theta = [];
+        this.L = [];
+        for(var d = 0; d < this.dim; ++d){
+            this.theta.push(0.0);
+            this.L.push(X[d]);
+        }
+        
+        for(var iter = 0; iter < this.iterations; ++iter){
+            var theta_delta = this.grad(X, Y, this.theta);
+            for(var d = 0; d < this.dim; ++d){
+                this.theta[d] = this.theta[d] - this.alpha * theta_delta[d];        
+            }
+            if(this.trace){
+                console.log('cost: ' + this.cost(X, Y, this.theta));
+            }
+        }
+        
+        var scores = this.transform(data);
+        
+        this.threshold = null;
+        
+        var N = data.length;
+        for(var i=0; i < N; ++i){
+            var label = data[i][this.dim-1];
+            if(label == 1){
+                if(this.threshold == null || this.threshold > scores[i]){
+                    this.threshold = scores[i];
+                }
+            }
+        }
+        
+        return {
+            theta: this.theta,
+            L: this.L,
+            cost: this.cost(X, Y, this.theta),
+            threshold: this.threshold,
+            config: {
+                alpha: this.alpha,
+                C: this.C,
+                iterations: this.iterations,
+                trace: this.trace,
+                sigma: this.sigma
+            }
+        }
+    };
+    
+    
+    
+    KernelSvm.prototype.smallStep = function(theta, d) {
+      return jsr.deltaStep(theta[d]);  
+    };
+    
+    KernelSvm.prototype.grad = function(X, Y, theta) {
+
+        var delta = [];
+        var cost_now = this.cost(X, Y, theta);
+        for(var d = 0; d < this.dim; ++d) {
+            var small_step = this.smallStep(theta, d);
+            var theta_d = theta[d];
+            theta[d] += small_step;
+            var cost_d = this.cost(X, Y, theta);
+            theta[d] = theta_d;
+            delta.push((cost_d - cost_now) / 2.0);
+        }
+        return delta;
+    };
+    
+    KernelSvm.prototype.cost = function(X, Y, theta) {
+        var N = X.length;
+        var sum = 0;
+        for(var i=0; i < N; ++i) {
+            var x_i = X[i];
+            var y_i = Y[i];
+            sum += this.C * (y_i * this.cost1(x_i, y_i, theta) + (1-y_i) * this.cost0(x_i, y_i, theta));
+        }
+        
+        sum += jsr.dotProduct(theta, theta) / 2.0;
+        
+        sum /= N;
+        return sum;
+    };
+    
+    KernelSvm.prototype.cost1 = function(x_i, y_i, theta) {
+        var z = jsr.gaussian(x_i, theta, this.L, this.sigma);
+        if(z > 1){
+            return 0;
+        }
+        return 1 - z;
+    };
+    
+    KernelSvm.prototype.cost0 = function(x_i, y_i, theta) {
+        var z = jsr.gaussian(x_i, theta, this.L, this.sigma);
+        if(z < -1) {
+            return 0;
+        }
+        return 1 + z;
+            
+    };
+    
+
+    
+    KernelSvm.prototype.transform = function(x) {
+        if(x[0].length){ // x is a matrix            
+            var predicted_array = [];
+            for(var i=0; i < x.length; ++i){
+                var predicted = this.transform(x[i]);
+                predicted_array.push(predicted);
+            }
+            return predicted_array;
+        }
+        
+        var x_i = [];
+        x_i.push(1.0);
+        for(var d = 1; d < this.dim; ++d){
+            x_i.push(x[d-1]);
+        }
+        
+        return jsr.gaussian(x_i, this.theta, this.L, this.sigma);  
+    };
+    
+    
+    jsr.KernelSvm = KernelSvm;
+    
     var BinarySvmClassifier = function(config){
-        this.classifier = new jsr.LinearSvm(config);
+        var config = config || {};
+        if(!config.kernel){
+            config.kernel = 'linear';
+        }
+        if(!config.sigma) {
+            config.sigma = 1.0;
+        }
+        this.kernel = config.kernel;
+        if(config.kernel == 'linear'){
+            this.classifier = new jsr.LinearSvm(config);
+        } else if(config.kernel == 'gaussian') {
+            this.classifier = new jsr.KernelSvm(config);
+        } else {
+            this.classifier = new jsr.LinearSvm(config);
+        }
     };
@@ -241,2 +443,3 @@
         
+        result.kernel = this.kernel;
         
@@ -248,3 +451,3 @@
     BinarySvmClassifier.prototype.transform = function(x) {
-        return this.classifier.transform(x) > this.threshold ? 1 : 0;
+        return this.classifier.transform(x) > this.classifier.threshold ? 1 : 0;
     };
@@ -257,13 +460,22 @@
         if(!config.alpha){
-            config.alpha = 0.001;
+            config.alpha = 0.01;
         }
         if(!config.iterations) {
-            config.iterations = 100;
+            config.iterations = 1000;
         }
         if(!config.C) {
-            config.C = C;
+            config.C = 5.0;
         }
+        if(!config.kernel) {
+            config.kernel = 'linear';
+        }
+        if(!config.sigma) {
+            config.sigma = 1.0;
+        }
+        
         this.alpha = config.alpha;
         this.C = config.C;
         this.iterations = config.iterations;
+        this.sigma = config.sigma;
+        this.kernel = config.kernel;
     };
@@ -298,7 +510,23 @@
             var c = this.classes[k];
-            this.classifiers[c] = new jsr.LinearSvm({
-                alpha: this.alpha,
-                C: this.C,
-                iterations: this.iterations
-            });
+            
+            if(this.kernel == 'linear'){
+                this.classifiers[c] = new jsr.LinearSvm({
+                    alpha: this.alpha,
+                    C: this.C,
+                    iterations: this.iterations
+                });
+            } else if(this.kernel == 'gaussian'){
+                this.classifiers[c] = new jsr.KernelSvm({
+                    alpha: this.alpha,
+                    C: this.C,
+                    iterations: this.iterations,
+                    sigma: this.sigma
+                });
+            } else {
+                this.classifiers[c] = new jsr.LinearSvm({
+                    alpha: this.alpha,
+                    C: this.C,
+                    iterations: this.iterations
+                });
+            }
             var data_c = [];
@@ -315,2 +543,4 @@ for(var i=0; i < N; ++i){
         }
+        
+        result.kernel = this.kernel;
         return result;
@@ -335,4 +565,7 @@ };
             var c = this.classes[k];
-            var prob_c = this.classifiers[c].transform(x) - this.classifiers[c].threshold;
-            if(max_prob < prob_c){
+            var prob_c;
+
+            
+            prob_c = this.classifiers[c].transform(x) - this.classifiers[c].threshold;
+            if(max_prob <= prob_c){
                 max_prob = prob_c;
@@ -339,0 +572,0 @@ best_c = c;

test/binary-classifier-spec.js

@@ -7,3 +7,3 @@ var expect    = require("chai").expect;
 
-  describe("solve the binary classification problem of iris data for which species class == Iris-virginica", function(){
+  describe("Use linear svm to solve the binary classification problem of iris data for which species class == Iris-virginica", function(){
        var svm = new jssvm.BinarySvmClassifier({
@@ -42,7 +42,46 @@ alpha: 0.01,
            var predicted = svm.transform(testingData[i]);
-           console.log("actual: " + testingData[i][4] + " predicted: " + predicted);
+           console.log("linear svm binary classifier testing: actual: " + testingData[i][4] + " predicted: " + predicted);
        }
     });
+    
+    describe("Use kernel svm to solve the binary classification problem of iris data for which species class == Iris-virginica", function(){
+       var svm = new jssvm.BinarySvmClassifier({
+           alpha: 0.01,
+           iterations: 1000,
+           C: 5.0,
+           trace: false,
+           kernel: 'gaussian'
+       });
+        
+       iris.shuffle();
+        
+       var trainingDataSize = Math.round(iris.rowCount * 0.9);
+       var trainingData = [];
+       var testingData = [];
+       for(var i=0; i < iris.rowCount ; ++i) {
+           var row = [];
+           row.push(iris.data[i][0]); // sepalLength;
+           row.push(iris.data[i][1]); // sepalWidth;
+           row.push(iris.data[i][2]); // petalLength;
+           row.push(iris.data[i][3]); // petalWidth;
+           row.push(iris.data[i][4] == "Iris-virginica" ? 1.0 : 0.0); // output which is 1 if species is Iris-virginica; 0 otherwise
+           if(i < trainingDataSize){
+                trainingData.push(row);
+           } else {
+               testingData.push(row);
+           }
+       }
+       
+        
+       var result = svm.fit(trainingData);
+        
+       console.log(result);
+        
+       for(var i=0; i < testingData.length; ++i){
+           var predicted = svm.transform(testingData[i]);
+           console.log("kernel svm binary classifier testing: actual: " + testingData[i][4] + " predicted: " + predicted);
+       }
+    });
 });
 
    

test/linear-svm-spec.js

@@ -41,3 +41,3 @@ var expect    = require("chai").expect;
            var probabilityOfSpeciesBeingIrisVirginica = svm.transform(testingData[i]);
-           console.log("actual: " + testingData[i][4] + " probability of being Iris-virginica: " + probabilityOfSpeciesBeingIrisVirginica);
+           console.log("linear svm test: actual: " + testingData[i][4] + " probability of being Iris-virginica: " + probabilityOfSpeciesBeingIrisVirginica);
        }
@@ -44,0 +44,0 @@

test/multi-class-classifier-spec.js

@@ -6,3 +6,3 @@ var expect    = require("chai").expect;
 describe("Test multi-class classification using logistic regression", function(){
-    describe("solve the multi-class classification problem of iris data", function(){
+    describe("Use linear svm to solve the multi-class classification problem of iris data", function(){
        var classifier = new jssvm.MultiClassSvmClassifier({
@@ -40,6 +40,44 @@ alpha: 0.001,
            var predicted = classifier.transform(testingData[i]);
-           console.log("svm prediction testing: actual: " + testingData[i][4] + " predicted: " + predicted);
+           console.log("linear svm multi-class prediction testing: actual: " + testingData[i][4] + " predicted: " + predicted);
        }
         
     });
+    
+    describe("Use kernel svm solve the multi-class classification problem of iris data", function(){
+       var classifier = new jssvm.MultiClassSvmClassifier({
+           alpha: 0.001,
+           iterations: 1000,
+           C: 5.0
+       });
+        
+       iris.shuffle();
+        
+       var trainingDataSize = Math.round(iris.rowCount * 0.9);
+       var trainingData = [];
+       var testingData = [];
+       for(var i=0; i < iris.rowCount ; ++i) {
+           var row = [];
+           row.push(iris.data[i][0]); // sepalLength;
+           row.push(iris.data[i][1]); // sepalWidth;
+           row.push(iris.data[i][2]); // petalLength;
+           row.push(iris.data[i][3]); // petalWidth;
+           row.push(iris.data[i][4]); // output is species
+           if(i < trainingDataSize){
+                trainingData.push(row);
+           } else {
+               testingData.push(row);
+           }
+       }
+       
+        
+       var result = classifier.fit(trainingData);
+        
+       console.log(result);
+        
+       for(var i=0; i < testingData.length; ++i){
+           var predicted = classifier.transform(testingData[i]);
+           console.log("kernel svm multi-class prediction testing: actual: " + testingData[i][4] + " predicted: " + predicted);
+       }
+        
+    });
 });

No alert changes

Improved metrics

Total package byte prevSize

40882

increased by59.11%

Number of package files

14

increased by7.69%

Lines of code

747

increased by66%

Number of lines in readme file

155

increased by32.48%

No dependency changes