GitHub - DrRoach/JS-Machine-Learning: Machine learning toolkit written in JavaScript

JS Machine Learning or JSML is a simple JavaScript machine learning toolkit. It's a small little pet project of mine and It's aim is to provide a range of different machine learning techniques which can easily be used either for educational use or real world problems.

View Demos

Hill Climbing

Hill climbing is a optimization technique used to find optimum values in large datasets. This technique can be applied to problems such as the travelling salesman issue more here.

Usage

    var ml = new MachineLearning();
    var data = [
        0,
        3,
        4,
        7,
        8,
        12,
        11,
        10,
        8,
        4,
        3,
        1
    ];
    var optima = ml.hillclimbing(data);

Bayes Predictions

Bayes theorem implementation that is used to calculate outputs given a set of inputs. This technique can be applied to making predictions such as choosing the winner of a football match more here.

Usage

    var ml = new MachineLearning();
    // Whether focus team was home or away
    var inputs = [["h", "a", "h", "a", "h", "h", "a", "h", "h", "a"]];
    // Whether focus team won, drew or last corresponding input
    var outputs = ["l", "l", "l", "l", "w", "d", "w", "l", "w", "d"];
    // Out inputs to use to make our prediction
    var next = ["h"];

    var bayes = ml.bayes(inputs, outputs, next);

Known Issue

The current bayes implementation only allows one input dimension. This is known and in the process of being fixed.

Neural Network

Neural networks can be trained to think by itself and spot patterns that we as humans may not be able to see.

Usage

    // Create network with 2 input nodes, 2 hidden nodes and one output node
    var nn = new NeuralNetwork(2, 2, 1);
    // Our training data. We have 2 inputs and 1 output
    var nnTraining = 
    [{
        inputs: [0,0],
        targets: [0]
    }, 
    {
        inputs: [1,0],
        targets: [1]
    },
    {
        inputs: [0,1],
        targets: [1]
    },
    {
        inputs: [1,1],
        targets: [0]
    }];
    // Train our neural network
    for (var i = 0; i < 100000; i++) {
        var rand = Math.floor(Math.random() * 4);
        var data = nnTraining[rand];
        nn.train(data.inputs, data.targets);
    }
    // Make our guess
    var guess = nn.feedforward([0, 1]);
    console.log(guess);

Currently Working On

Fixing TSP implementation to give more accurate results
Change Bayes implementation to allow for more than one set of input data
Allowing more than one hidden neuron layer
Simulated Annealing tool
k-NearestNeighbour tool
Evolutionary Algorithms