Simple Neural Network in Go

simple_nn.go

/* Simple neural net with one hidden layer consisting of one neuron */
/* Inspired by https://medium.com/technology-invention-and-more/how-to-build-a-simple-neural-network-in-9-lines-of-python-code-cc8f23647ca1 */

package main

import (
	"fmt"
	"math/rand"
	"math"
)

type activation func(float64) float64

func sigmoid(x float64) float64 {
	return 1 / (1+math.Exp(x * (-1)))
}

func sigmoid_d(x float64) float64 {
	return math.Exp(x) / math.Pow((math.Exp(x) + 1), 2.0)
}

func activate(x, w []float64, fn activation) float64 {
	raw := 0.0
	for j := 0; j < len(x); j += 1 {
		raw += x[j] * w[j]
	}
	return fn(raw)
}

func totalMseLoss(X [][]float64, y, w []float64) float64 {
	loss := 0.0
	for i := 0; i < len(X); i += 1 {
		loss += math.Pow(y[i] - activate(X[i], w, sigmoid), 2.0)
	}
	return loss / float64(len(X))
}

func info(X [][]float64, y, w0 []float64, epoch int) {
	fmt.Println("Iteration", epoch + 1)
	fmt.Println("Loss: ", totalMseLoss(X, y, w0))
	for i := 0; i < len(X); i += 1 {
		fmt.Printf("%.4f ", activate(X[i], w0, sigmoid))
		if i == len(X) - 1 {
			fmt.Println()
		}
	}
	for i := 0; i < len(w0); i += 1 {
		fmt.Printf("w%d %.4f ", i, w0[i])
		if i == len(w0) - 1 {
			fmt.Println()
		}
	}
	fmt.Println("------")
}

func main() {
	rate := 0.1
	X := [][]float64{
		[]float64{0.0, 0.0, 1.0},
		[]float64{0.0, 1.0, 1.0},
		[]float64{1.0, 0.0, 1.0},
		[]float64{1.0, 1.0, 1.0},
	}
	y := []float64{0.0, 0.0, 1.0, 1.0}

	w0 := make([]float64, len(X[0]))
	for i := 0; i < len(w0); i += 1 {
		w0[i] = 2*rand.Float64() - 1
	}

	for epoch := 0; epoch < 1000; epoch += 1 {
		for i := 0; i < len(X); i += 1 {
			out := activate(X[i], w0, sigmoid)
			error := y[i] - out
			for j := 0; j < len(X[0]); j += 1 {
				// See (22) in http://www.idi.ntnu.no/~keithd/classes/advai/lectures/backprop.pdf
				w0[j] += rate * X[i][j] * error * sigmoid_d(out)
			}
		}
		
		if epoch % 100 == 0 {
			info(X, y, w0, epoch)
		}
	}
}