A Bare-bones Machine Learning Library in Go

This Go library provides an API which allows for basic machine learning capability. This library contains no external dependencies, is completely go-based, and fully implements a neural network.

I followed along with a C++ version while making this library. Links to that version are below.

Links: Repository - Videos

Documentation / Usage

The full API documentation is available on pkg.go.dev.

Installation

go get -u github.com/justincpresley/go-bareml

Or use your favorite golang vendoring tool.

Example

package main

import (
	"fmt"

	"github.com/justincpresley/go-bareml"
)

func main() {
	var input = []float64{0.2, 0.5, 0.1}
	var target = []float64{0.2, 0.5}

	// The first int must be len(input) and the last int must be len(target)
	var topology = []int{3, 2, 10, 2}

	var (
		learningRate float64 = 0.05
		momentum     float64 = 1
		bias         float64 = 1
	)

	neuralnet := bareml.NewNeuralNetwork(topology)

	fmt.Printf("Begin Training...\n")
	for i := 0; i < 1000; i++ {
		neuralnet.Train(input, target, bias, learningRate, momentum)
		fmt.Printf("Epoch: %d | Error %f\n", i, neuralnet.TotalError())
	}
}

Notes

This was a learning project for me to get comfortable with Go. I originally aimed to improve the design and add additional features / models. However, I am now unable to maintain this library due to more pressing work in networking and cybersecurity. Feel free to fork the repository and build off it though!