Concept

Just another NN-learning library for personal usage and education

Usage

Importing

On Linux after cloning you can already import this lib (src/src.ts file) to your TS project, but to import to JS project you need to build TS code. npm install will install devDependencies and compile TS code (example and src) into /build folder

Functionality

After importing lib as NN you can:

create NN with specific layout like this:

        const nn = new NeuralNetwork(1, 2, 1); //makes NN with 3 layers with 1, 2 and 1 neuron respectively

get output of NN via feedforward():

        const out = nn.feedforward(data);

Note that input data should match first layer's size or exception will be thrown

learn dataset with train():

        const dataset = [[input1, output1], [input2, output2], [input3, output3],...];
        nn.train(...dataset);

Note that dimensions of input and output should match first and last layer's size respectively

dynamically change NN's structure with addNeuron()/addLayer() and removeNeuron()/removeLayer() functions
import/export NN from file with saveTo() and loadFrom()

Other available function is described in Docs.md

Tests

Tests can be run with npm test (after npm install was once called) which will lint and run test scripts from /tests folder Tests check wheather code works for simplest case of NNs and some features. More information about tests is available Docs.md

Examples