LearningML

Learningml is an open-source machine learning library for training models, currently only supports ANNs.

Works in the browser or in node extremely flexible, with a focus on having minimal dependencies. Easy to work with made from the ground up in javascript, manipulate objects in an intuative way. Constantly expanding Plans to add gpu acceleration, alternative neural network architectures, and more!

Examples

all examples can be viewed in the examples folder.

npm i learningml

to install the package.

const lml = require('learningml');

to use in your project

you want your inputs to be flattened so you can put it into vector form.

const input = lml.Vector.vector(x1, x2, x3, ..., xn)

this is how you can create a valide vector with the x's being the values that make up an input value.

const output = input.apply(x => x/2)

this will make a new vector with values half that of the input

const prenetwork = lml.Network.network(n,n,
	[
		lml.Layer.layer(lml.Layer.simple(n), lml.Layer.relu),
		lml.Layer.layer(lml.Layer.simple(n), lml.Layer.relu),
		lml.Layer.layer(lml.Layer.simple(n), lml.Layer.relu),
	]
);

next we need to define our network which takes in an input size, an output size, and a array of layers. a simple layer is simply a fully connected layer, and relu is the relu activation function.

let network = lml.Network.generate(prenetwork)

we generate the network to finalize it(reccomended to use let instead of const here).

const result = lml.F.feedforward(input, network)

to use the network with a given input.

network = lml.Trainer.train(
	inputs,
	outputs,
	network,
	epochs,
	lml.Cost.leastsquares,
	lml.Optimizer.linear(.07)
)

you train the network by giving it an array of input and output vectors where:

(inputs[i], outputs[i]) //correspond to an input/output pair

once the network successfully trains you can use the model with the feedforward function.

Contribution guide

want to contribute? thank you for your interest. there is no standard contribution workflow so just submit your issues and PR and I will look at them. Thank you!