cscの作法 その32 ニューラルネットワーク

概要

cscの作法、調べてみた。
tensorflow.jsでxorを学習して、バイアスとウェイトを取り出して、c#で推論してみた。

写真

サンプルコード

using System;
using System.Windows.Forms;
using System.Drawing;

class form1: Form {
	form1() {
		Text = "nn";
		ClientSize = new Size(200, 200);
		double[][] w1 = {
			new double[] {-0.8971094, 1.4029436, -1.3633214, -0.775598, 1.2531579, 1.0529767, 1.3460118, -1.5597397},
			new double[] {1.57735, -1.1888924, 1.6184694, -0.5399966, -0.9783241, 0.9480417, 0.9573314, -1.1738404}
		};
		double[] b1 = {0.428151, 0.7863605, 0.8015493, 0.8869765, 0.6552361, 0.3099377, 0.1825127, 0.0495548};
		double[][] w2 = {
			new double[] {-1.7588012},
			new double[] {-1.5193572},
			new double[] {-1.1239196},
			new double[] {0.8726832},
			new double[] {-1.7272931},
			new double[] {1.4907871},
			new double[] {1.1091278},
			new double[] {-0.9800876}
		};
		double[] b2 = {0.1271933};
		string res = "";
		for (int i = 0; i < 4; i++)
		{
			int k;
			int l;
			double[] x = {0, 0};
			double[] y = {0};
			double[] h = {0, 0, 0, 0, 0, 0, 0, 0};
			if (i == 0)
			{
				x[0] = 1.0;
				x[1] = 0;
				res += "1, 0 = ";
			}
			if (i == 1)
			{
				x[0] = 0;
				x[1] = 1.0;
				res += "0, 1 = ";
			}
			if (i == 2)
			{
				x[0] = 1.0;
				x[1] = 1.0;
				res += "1, 1 = ";
			}
			if (i == 3)
			{
				x[0] = 0;
				x[1] = 0;
				res += "0, 0 = ";
			}
			for (k = 0; k < 2; k++)
			{
				for (l = 0; l < 8; l++)
				{
					h[l] += x[k] * w1[k][l];
				}
			}
			for (l = 0; l < 8; l++)
			{
				h[l] += b1[l];
				h[l] = tanh(h[l]);
			}
			for (k = 0; k < 8; k++)
			{
				for (l = 0; l < 1; l++)
				{
					y[l] += h[k] * w2[k][l];
				}
			}
			for (k = 0; k < 1; k++)
			{
				y[k] += b2[k];
				y[k] = sigmoid(y[k]);
			}
			res += y[0].ToString() + "\r\n";
		}
		MessageBox.Show(res);
	}
	double tanh(double x) {
		double a = Math.Exp(x);
		double b = Math.Exp(-x);
		return (a - b) / (a + b);
	}
	double sigmoid(double x) {
		return 1.0 / (1.0 + Math.Exp(-x));
	}
	[STAThread]
	public static void Main() {
		Application.Run(new form1());
	}
}

以上。