GitHub - LuKks/neural-go: Genetic Neural Networks

Genetic Neural Networks

package main

import (
  "fmt"
  "github.com/lukks/neural-go/v3"
)

func main() {
  xor := neural.NewNeural([]*neural.Layer{
    {Inputs: 2, Units: 16},
    {Units: 16},
    {Units: 1},
  })

  for i := 0; i <= 5000; i++ {
    loss := xor.Learns([][][]float64{
      {{0, 0}, {0}},
      {{1, 0}, {1}},
      {{0, 1}, {1}},
      {{1, 1}, {0}},
    })

    if i%1000 == 0 {
      fmt.Printf("iter %v, loss %f\n", i, loss)
    }
  }

  fmt.Printf("think some values:\n")
  fmt.Printf("0, 0 [0] -> %f\n", xor.Think([]float64{0, 0}))
  fmt.Printf("1, 0 [1] -> %f\n", xor.Think([]float64{1, 0}))
  fmt.Printf("0, 1 [1] -> %f\n", xor.Think([]float64{0, 1}))
  fmt.Printf("1, 1 [0] -> %f\n", xor.Think([]float64{1, 1}))
}

Install latest version

go get github.com/lukks/neural-go/v3

Also find versions on releases. The changes from v2 to v3 were just for go mod versioning.

Features

Range

Set a range of values for every input and output.
So you use your values as you know but the neural get it in raw activation.
Check examples/rgb.go for usage example.

Customizable

Set different activations, rates, momentums, etc at layer level.

Activation: linear, sigmoid (default), tanh and relu
Learning Rate
Optimizer by Momentum
Loss: for output layer, only mse for now
Range: for input and output layer

Check examples/layers.go for complete example.

Genetics

Clone, mutate and crossover neurons, layers and neurals.
The Evolve method internally uses these methods to put this very easy.
Check examples/evolve.go but it's optional, not always need to use genetics.

Utils

There are several useful methods: Export, Import, Reset, ToFile, FromFile, etc.
Check the documentation here.

Description

From my previous neural-amxx.

Examples

Basic XOR examples/xor.go
RGB brightness examples/rgb.go
Genetics examples/evolve.go
Layer configs examples/layers.go
Persist examples/persist.go