ml-svm 2.1.1 | Documentation

ml-svm

2.1.1

SVM ▸
- Static members
- .load
- Instance members
- #train
- #predict
- #margin
- #supportVectors
- #toJSON

Need help reading this?

SVM

src/svm.js

Simplified version of the Sequential Minimal Optimization algorithm for training support vector machines

new SVM(options: {Object})

Parameters

options ({Object}) SVM options

Name	Description
options.C `[Number]` (default `1`)	regularization parameter
options.tol `[Number]` (default `1e-4`)	numerical tolerance
options.alphaTol `[Number]` (default `1e-6`)	alpha tolerance, threshold to decide support vectors
options.maxPasses `[Number]` (default `10`)	max number of times to iterate over alphas without changing
options.maxIterations `[Number]` (default `10000`)	max number of iterations
options.kernel `[String]` (default `linear`)	the kind of kernel. List of kernels
options.random `[Function]` (default `Math.random`)	custom random number generator

Static Members

▸ load(model)

Create a SVM instance from a saved model

load(model: Object): SVM

Parameters

model (Object) Object such as returned by a trained SVM instance with toJSON

Returns

SVM: Instance of svm classifier

Instance Members

▸ train(features, labels)

Train the SVM model

train(features: Array<Array<number>>, labels: Array<number>)

Parameters

features (Array<Array<number>>) training data features

labels (Array<number>) training data labels in the domain {1,-1}

▸ predict(features)

Predict the classification outcome of a trained svm given one or several observations' features.

predict(features: Array): (Array<Number> | Number)

Parameters

features (Array) The observation(s)' features

Returns

(Array<Number> | Number): An array of {-1, 1} if several observations are given or a number if one observation is given

▸ margin(features)

Returns the margin of one or several observations given its features

margin(features: (Array<Array<Number>> | Array<Number>)): (Number | Array)

Parameters

features ((Array<Array<Number>> | Array<Number>)) Features from on or several observations.

Returns

(Number | Array): The computed margin. Is an Array if several observations' features given, or a Number if only one observation's features given

▸ supportVectors()

Get support vectors indexes of the trained classifier. WARINNG: this method does not work for svm instances created from load if linear kernel

supportVectors(): Array<Number>

Returns

Array<Number>: The indices in the training vector of the support vectors

▸ toJSON()

Export the minimal object that enables to reload the model

toJSON(): Object

Returns

Object: Model object that can be reused with load