node-svm

Support Vector Machine (SVM) library for nodejs & io.js .

Support Vector Machines

Wikipedia :

Support vector machines are supervised learning models that analyze data and recognize patterns. A special property is that they simultaneously minimize the empirical classification error and maximize the geometric margin; hence they are also known as maximum margin classifiers.

Installation

npm install --save node-svm

Quick start

If you are not familiar with SVM I highly recommend this guide.

Here's an example of using node-svm to approximate the XOR function :

varsvm=require('node-svm');varxor=[[[0,0],0],[[0,1],1],[[1,0],1],[[1,1],0]];// initialize a new predictorvarclf=newsvm.CSVC();clf.train(xor).done(function(){// predict thingsxor.forEach(function(ex){varprediction=clf.predictSync(ex[0]);console.log('%d XOR %d => %d',ex[0][0],ex[0][1],prediction);});});/******** CONSOLE ******** 0 XOR 0 => 0 0 XOR 1 => 1 1 XOR 0 => 1 1 XOR 1 => 0 */

More examples are available here.

Note: There's no reason to use SVM to figure out XOR BTW...

API

Classifiers

Possible classifiers are:

Classifier	Type	Params	Initialization
C_SVC	multi-class classifier	c	= new svm.CSVC(opts)
NU_SVC	multi-class classifier	nu	= new svm.NuSVC(opts)
ONE_CLASS	one-class classifier	nu	= new svm.OneClassSVM(opts)
EPSILON_SVR	regression	c, epsilon	= new svm.EpsilonSVR(opts)
NU_SVR	regression	c, nu	= new svm.NuSVR(opts)

Kernels

Possible kernels are:

Kernel	Parameters
LINEAR	No parameter
POLY	degree, gamma, r
RBF	gamma
SIGMOID	gamma, r

Parameters and options

Possible parameters/options are:

Name	Default value(s)	Description
svmType	C_SVC	Used classifier
kernelType	RBF	Used kernel
c	[0.01,0.125,0.5,1,2]	Cost for C_SVC, EPSILON_SVR and NU_SVR. Can be a Number or an Array of numbers
nu	[0.01,0.125,0.5,1]	For NU_SVC, ONE_CLASS and NU_SVR. Can be a Number or an Array of numbers
epsilon	[0.01,0.125,0.5,1]	For EPSILON_SVR. Can be a Number or an Array of numbers
degree	[2,3,4]	For POLY kernel. Can be a Number or an Array of numbers
gamma	[0.001,0.01,0.5]	For POLY, RBF and SIGMOID kernels. Can be a Number or an Array of numbers
r	[0.125,0.5,0,1]	For POLY and SIGMOID kernels. Can be a Number or an Array of numbers
kFold	4	k parameter for k-fold cross validation. k must be >= 1. If k===1 then entire dataset is use for both testing and training.
normalize	true	Whether to use mean normalization during data pre-processing
reduce	true	Whether to use PCA to reduce dataset's dimensions during data pre-processing
retainedVariance	0.99	Define the acceptable impact on data integrity (require reduce to be true)
eps	1e-3	Tolerance of termination criterion
cacheSize	200	Cache size in MB.
shrinking	true	Whether to use the shrinking heuristics
probability	false	Whether to train a SVC or SVR model for probability estimates

The example below shows how to use them:

varsvm=require('node-svm');varclf=newsvm.SVM({svmType: 'C_SVC',c: [0.03125,0.125,0.5,2,8],// kernels parameterskernelType: 'RBF',gamma: [0.03125,0.125,0.5,2,8],// training optionskFold: 4,normalize: true,reduce: true,retainedVariance: 0.99,eps: 1e-3,cacheSize: 200,shrinking : true,probability : false});

Notes :

• You can override default values by creating a .nodesvmrc file (JSON) at the root of your project.
• If at least one parameter has multiple values, node-svm will go through all possible combinations to see which one gives the best results (it performs grid-search to maximize f-score for classification and minimize Mean Squared Error for regression).

##Training

SVMs can be trained using svm#train(dataset) method.

Pseudo code :

varclf=newsvm.SVM(options);clf.train(dataset).progress(function(rate){// ...}).spread(function(trainedModel,trainingReport){// ...});

Notes :

• trainedModel can be used to restore the predictor later (see this example for more information).
• trainingReport contains information about predictor's accuracy (such as MSE, precison, recall, fscore, retained variance etc.)

Prediction

Once trained, you can use the classifier object to predict values for new inputs. You can do so :

• Synchronously using clf#predictSync(inputs)
• Asynchronously using clf#predict(inputs).then(function(predicted){ ... });

If you enabled probabilities during initialization you can also predict probabilities for each class :

• Synchronously using clf#predictProbabilitiesSync(inputs).
• Asynchronously using clf#predictProbabilities(inputs).then(function(probabilities){ ... }).

Note : inputs must be a 1d array of numbers

Model evaluation

Once the predictor is trained it can be evaluated against a test set.

Pseudo code :

varsvm=require('node-svm');varclf=newsvm.SVM(options);svm.read(trainFile).then(function(dataset){returnclf.train(dataset);}).then(function(trainedModel,trainingReport){returnsvm.read(testFile);}).then(function(testset){returnclf.evaluate(testset);});.done(function(report){console.log(report);});

CLI

node-svm comes with a build-in Command Line Interpreter.

To use it you have to install node-svm globally using npm install -g node-svm.

See $ node-svm -h for complete command line reference.

help

$ node-svm help [<command>]

Display help information about node-svm

train

$ node-svm train <dataset file> [<where to save the prediction model>] [<options>]

Train a new model with given data set

Note: use $ node-svm train <dataset file> -i to set parameters values dynamically.

evaluate

$ node-svm evaluate <model file><testset file> [<options>]

Evaluate model's accuracy against a test set

How it work

node-svm uses the official libsvm C++ library, version 3.20.

For more information see also :

• libsvm web site
• Chih-Chung Chang and Chih-Jen Lin, LIBSVM : a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011.
• Wikipedia article about SVM
• node addons

Contributions

Feel free to fork and improve/enhance node-svm in any way your want.

If you feel that the community will benefit from your changes, please send a pull request :

• Fork the project.
• Make your feature addition or bug fix.
• Add documentation if necessary.
• Add tests for it. This is important so I don't break it in a future version unintentionally (run grunt or npm test).
• Send a pull request to the develop branch.

#FAQ ###Segmentation fault Q : Node returns 'segmentation fault' error during training. What's going on?

A1 : Your dataset is empty or its format is incorrect.

A2 : Your dataset is too big.

###Difference between nu-SVC and C-SVC Q : What is the difference between nu-SVC and C-SVC?

A : Answer here

###Other questions

• Take a look at libsvm's FAQ.
• Create an issue

License

MIT