cross-validation

Utility library to do cross validation with supervised classifiers.

Cross-validation methods:

• k-fold
• leave-p-out

API documentation.

A list of the mljs supervised classifiers is available here in the supervised learning section, but you could also use your own. Cross validations methods return a ConfusionMatrix (https://github.com/mljs/confusion-matrix) that can be used to calculate metrics on your classification result.

Installation

npm i -s ml-cross-validation

Example using a ml classification library

constcrossValidation=require('ml-cross-validation');constKNN=require('ml-knn');constdataset=[[0,0,0],[0,1,1],[1,1,0],[2,2,2],[1,2,2],[2,1,2]];constlabels=[0,0,0,1,1,1];constconfusionMatrix=crossValidation.leaveOneOut(KNN,dataSet,labels);constaccuracy=confusionMatrix.getAccuracy();

Example using a classifier with its own specific API

If you have a library that does not comply with the ML Classifier conventions, you can use can use a callback to perform the classification. The callback will take the train features and labels, and the test features. The callback shoud return the array of predicted labels.

constcrossValidation=require('ml-cross-validation');constKNN=require('ml-knn');constdataset=[[0,0,0],[0,1,1],[1,1,0],[2,2,2],[1,2,2],[2,1,2]];constlabels=[0,0,0,1,1,1];constconfusionMatrix=crossValidation.leaveOneOut(dataSet,labels,function(trainFeatures,trainLabels,testFeatures){constknn=newKNN(trainFeatures,trainLabels);returnknn.predict(testFeatures);});constaccuracy=confusionMatrix.getAccuracy();

ML classifier API conventions

You can write your classification library so that it can be used with ml-cross-validation as described in here For that, your classification library must implement

• A constructor. The constructor can be passed options as a single argument.
• A train method. The train method is passed the data as a first argument and the labels as a second.
• A predict method. The predict method is passed test data and should return a predicted label.

Example

classMyClassifier{constructor(options){this.options=options;}train(data,labels){// Create your model}predict(testData){// Apply your model and return predicted labelreturnprediction;}}