nodeml

nodeml

Machine Learning Framework for Node

Summary

• Feature Selection
  • nodeml.feature.tfidf: tfidf
• Classification
  • nodeml.Bayes: Bayes
  • nodeml.kNN: k-Nearest Neighbor
  • nodeml.CNN: Convolutional Neural Network (CNN)
• Clustering
  • nodeml.kMeans: k-Means
• Recommendation
  • nodeml.CF: User based Collaborative Filtering
• Evaluation
  • nodeml.accuracy: Precision, Recall, F-Measure, Accuracy
  • nodeml.ndcg: NDCG

Installation

installation on your project

npm install --save nodeml

use example

const {Bayes} = require('nodeml');
let bayes = new Bayes();

bayes.train({'fun': 3, 'couple': 1}, 'comedy');
bayes.train({'couple': 1, 'fast': 1, 'fun': 3}, 'comedy');
bayes.train({'fast': 3, 'furious': 2, 'shoot': 2}, 'action');
bayes.train({'furious': 2, 'shoot': 4, 'fun': 1}, 'action');
bayes.train({'fly': 2, 'fast': 3, 'shoot': 2, 'love': 1}, 'action');

let result = bayes.test({'fun': 3, 'fast': 3, 'shoot': 2});
console.log(result); // this print {answer: , score: }

Document

nodeml.sample

Sample dataset for test

const {sample} = require('nodeml');

// bbc: Function() => { dataset: [ {} , ... ], labels: [ ... ] }
// bbc news dataset, sparse matrix
const bbc = sample.bbc();

// yeast: Function() => { dataset: [ [] , ... ], labels: [ ... ] }
// yeast dataset, array data
const yeast = sample.yeast();

// iris: Function() => { dataset: [ [] , ... ], labels: [ ... ] }
// iris dataset, array data
const iris = sample.iris();

// movie: Function() => [{ movie_id: '1', user_id: '97', rating: '5', like: '17' }, ...]
// movie dataset, array data
const movie = sample.movie();

---

nodeml.Bayes

Naive Bayes classifier

const {Bayes} = require('nodeml');
let bayes = new Bayes(); // this is bayes classfier

train: Function(data, label) => model

training bayes classifier

bayes.train([0.2, 0.5, 0.7, 0.4], 1);       
bayes.train({ 'my': 20, 'home': 30 }, 1);   

// training bulk
bayes.train([[2, 5,], [2, 1,]], [1, 2]);    
bayes.train([{}, {}], [1, 2]);              

test: Function(data) => { answer: string, score: {} }

classify document

let result = bayes.test([2, 5, 1, 4]);
let result = bayes.test({'fun': 3, 'fast': 3, 'shoot': 2});

getModel: Function () => model

get trained result

let model = bayes.getModel();
let str = JSON.stringify(model);

setModel: Function (model)

set pre-trained

bayes.setModel(JSON.parse(str));

---

nodeml.kNN

k-Nearest Neighbor Classifier

const {kNN} = require('nodeml');
let knn = new kNN();

train: Function(dataset, labels) => model

training

knn.train([0.2, 0.5, 0.7, 0.4], 1);       
knn.train({ 'my': 20, 'home': 30 }, 1);   

// training bulk
knn.train([[2, 5,], [2, 1,]], [1, 2]);    
knn.train([{ 'my': 20, 'home': 30 }, { 'my': 5, 'home': 10 }], [1, 2]);              

test: Function(dataset, k) => [ class1, class2, class1 ]

classify document (default k is 3)

let result = knn.test([2, 5, 1, 4]);
let result = knn.test({'fun': 3, 'fast': 3, 'shoot': 2}, 5);

getModel: Function () => model

get trained result

let model = knn.getModel();
let str = JSON.stringify(model);

setModel: Function (model)

set pre-trained

knn.setModel(JSON.parse(str));

---

nodeml.CNN

Convolutional Neural Network, based convnetjs

const {CNN} = require('nodeml');
let cnn = new CNN();

configure: Function (options)

options object refer trainer option at convnetjs

cnn.configure({learning_rate: 0.1, momentum: 0.001, batch_size: 5, l2_decay: 0.0001});

setModel: Function (layer or model)

layer refer at convnetjs

var layer = [];
layer.push({type: 'input', out_sx: 1, out_sy: 1, out_depth: 8});
layer.push({type: 'svm', num_classes: 10});

cnn.makeLayer(layer);

// set pre-trained
cnn.setModel(JSON.parse(str));

train: Function (data, label)

cnn.train([0.2, 0.5, 0.7, 0.4], 1);       
cnn.train({ 'my': 20, 'home': 30 }, 1);   

// training bulk
cnn.train([[2, 5,], [2, 1,]], [1, 2]);    
cnn.train([{}, {}], [1, 2]);   

test: Function(data) => { answer: string, score: {} }

classify document

let result = cnn.test([2, 5, 1, 4]);
let result = cnn.test({'fun': 3, 'fast': 3, 'shoot': 2});

getModel: Function () => model

get trained result

let model = cnn.getModel();
let str = JSON.stringify(model);

---

nodeml.kMeans

k-Means Clustering

const {kMeans} = require('nodeml');
let kmeans = new kMeans();

train: Function(dataset, options) => model

training

kmeans.train([[2, 5,], [2, 1,]], {
    k: 10, dm: 0.00001, iter: 100,  
    proc: (iter, j, d)=> { console.log(iter, j, d); }
});

options	description	type	default
init	cluster initialize function: random, fuzzy (preparing)	string	'random'
k	number of cluster	integer	3
dm	distortion measure	float	0.00
iter	maximum iteration	integer	unlimited
labels	supervised learning, if labels exists, detect k automatically	array	null
proc	process handler	function	null

test: Function(dataset) => [ class1, class2, class1 ]

classify document (default k is 3)

let result = kmeans.test([[2, 5,], [2, 1,]]);

getModel: Function () => model

get trained result

let model = kmeans.getModel();
let str = JSON.stringify(model);

setModel: Function (model)

set pre-trained

kmeans.setModel(JSON.parse(str));

---

nodeml.CF

Collaborative Filtering Function

const {CF, evaluation} = require('../index');

let train = [[1, 1, 2], [1, 2, 2], [1, 4, 5], [2, 3, 2],
    [2, 5, 1], [3, 1, 2], [3, 2, 3], [3, 3, 3]];
let test = [[3, 4, 1]];

const cf = new CF();
cf.train(train);
let gt = cf.gt(test);
let result = cf.recommendGT(gt, 1);

let ndcg = evaluation.ndcg(gt, result);

console.log(gt);
console.log(result);
console.log(ndcg);

train: Function

---

nodeml.evaluate

accuracy: Function (gt, result) => {precision, recall, f-measure, accuracy}

let {evaluate} = require('nodeml');

let original = [1, 2, 1, 1, 3]; // original label
let result = [1, 1, 2, 1, 3]; // train result label

// exec evaluate, this contains accuracy, micro/macro precision/recall/f-measure
let accuracy = evaluate.accuracy(original, result);

ndcg: Function (gt, result) => 0 ~ 1 ndcg value

let {CF, evaluate} = require('nodeml');
const cf = new CF();
let gt = cf.gt(test, 'user_id', 'movie_id', 'rating');

let result = cf.recommandToUsers(users, 40);

let ndcg = evaluation.ndcg(gt, result);