https://github.com/troublete/go-qndnn

quick 'n' dirty neural network (for practical use)
https://github.com/troublete/go-qndnn

ai deep-neural-networks go machine-learning relu sigmoid tanh

Last synced: 6 days ago
JSON representation

quick 'n' dirty neural network (for practical use)

• Host: GitHub
• URL: https://github.com/troublete/go-qndnn
• Owner: troublete
• License: mit
• Created: 2024-11-09T18:54:25.000Z (2 months ago)
• Default Branch: master
• Last Pushed: 2024-11-23T19:57:21.000Z (about 2 months ago)
• Last Synced: 2024-11-23T20:30:55.116Z (about 2 months ago)
• Topics: ai, deep-neural-networks, go, machine-learning, relu, sigmoid, tanh
• Language: Go
• Homepage:
• Size: 16.6 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# go-qndnn

> quick 'n' dirty neural network (for practical use)

## Introduction

This package contains a simple Go implementation for neural networks; for practical everyday-use in common use-cases. It

is neither heavily optimized to be the best package around, nor does this package contain an exhaustive variety of

mathematical functions. It supports Sigmoid, Tanh and ReLU. It leverages Go primitives.

```go

nn := qndnn.NewNeuralNet(nil, 4, 3, 3, 1) // sigmoid is default; input (4), hidden1 (3), hidden2 (3), output (1)

// qndnn.NewNeuralNet(qndnn.WithRelu(), 4, 3, 3, 1) // – to use with relu

// qndnn.NewNeuralNet(qndnn.WithTanh(), 4, 3, 3, 1) // - to use with tanh

// to retrieve output with input values

out, err := nn.Output([]float64{1, 2, 3, 4})

// to train on expectations

err = nn.Train(

	[]Expectation{

        {

            Input: []float64{1, 2, 3, 4},

            Output: []float64{.42},

        },

    }, 

	0.01, // learning rate

	qndnn.RoundStrategy(1000), // train for 1000 rounds; other options include ThresholdStrategy (see examples)

)

serializedBase64, err := nn.Serialize() // to serialize net (weights, biases)

nn, err = NewNeuralNetFromSerialized(nil, serializedBase64) // deserialize serialized net into usable structure; initialized with sigmoid 

//nn, err = NewNeuralNetFromSerialized(qndnn.WithRelu(), serializedBase64) // - to initialize with relu

//nn, err = NewNeuralNetFromSerialized(qndnn.WithTanh(), serializedBase64) // - to initialize with tanh

```