static void Main(string[] args)
{
NN nn = new NN();
nn.Run();
Console.ReadKey();
}
private static void TrainAll()
{
//-------------------primas NN-------------------
// Tikslas: nustatyti optimaliausia diena pamokai iterpti
// Duomenys: visos savaites klases, kai langas = 0, uzimta pamoka = 1, tokia pati pamoka = 0.5
// Rezultatai: diena nuo 1 iki 5 i kuria reikia ideti pamoka
//
List <NN> firstNets = new List <NN>(); // NN tinklu sarasas
int[] firstLayers = new int[] { 40, 20, 10, 1 }; //sluoksniu sistema
NN firstNet = new NN(firstLayers);
TrainFirst(ref firstNets, firstLayers);
//-------------------antras NN-------------------
// Tikslas: nustatyti optimaliausia langa pamokai iterpti
// Duomenys: vienos dienos pamokos, kai langas = 0, uzimta pamoka = 1, tokia pati pamoka = 0.5
// Rezultatai: langas nuo 1 iki 8 i kuri reikia ideti pamoka
//
List <NN> secondNets = new List <NN>(); // NN tinklu sarasas
int[] secondLayers = new int[] { 8, 10, 10, 1 }; //sluoksniu sistema
NN secondNet = new NN(secondLayers);
TrainSecond(ref secondNets, secondLayers);
}
public int CompareTo(NN other) //skirtas ivertinti ar NN yra geresnis uz kita
{
if (other == null)
{
return(1);
}
if (fitness > other.fitness)
{
return(1);
}
else if (fitness <= other.fitness)
{
return(-1);
}
else
{
return(0);
}
//reiksme 1: esamas NN yra geresnis arba kitas NN neaprasytas
//reiksme 0: klaida
//reiksme -1: esamas NN yra blogesnis uz kita NN
} // CompareTo()
public static float LaunchFirst(NN net, int[] layers, float[] chosen)
{
StreamReader sk = new StreamReader(@"weightValues1.txt", System.Text.Encoding.GetEncoding(1257));
net.ReadWeights(sk);
sk.Close();
float output = ((net.FeedForward(chosen)[0]) + 1f) * 2.5f; //gaunamas outputas is neural network
Console.WriteLine("First AI result: {0}", output);
List <float> possibilities = new List <float>(); //galimos vietos ikelti pamoka
for (int i = 0; i < 5; i++)
{
if (chosen[i * 3 + 2] == 0f || chosen[i * 3] + chosen[i * 3 + 1] == 8f)
{
possibilities.Add(i);
}
if (chosen[i * 3 + 2] == 0f && chosen[i * 3] + chosen[i * 3 + 1] == 8f)
{
possibilities.Add(i);
}
}
float nearest = ClosestTo(possibilities, output);
return(nearest);
}
public Form1()
{
InitializeComponent();
nn = new NN(2, 1, 1, 4, 1);
pool = new GP(poolSize, nn.weightCount());
for (byte b = 0; b < 4; b++)
{
bits[b, 0] = GetBit(b, 0) ? 1 : -1;
bits[b, 1] = GetBit(b, 1) ? 1 : -1;
}
}
static void InitNN(ref List <NN> nets, ref int populationSize, int[] layers) //NN apsirasymas
{
if (populationSize % 2 != 0) //jeigu nelyginis, paverciamas lyginiu
{
populationSize--;
}
for (int i = 0; i < populationSize; i++) //kiekvienam NN
{
NN net = new NN(layers); //naujas net su layers
net.Mutate(); //mutuojamas
nets.Add(net); //pridedamas prie bendro NN saraso
}
}
} // NN()
public NN(NN copyNetwork) // Constructor, perkopijuoja kita NN
{
this.layers = new int[copyNetwork.layers.Length];
for (int i = 0; i < copyNetwork.layers.Length; i++)
{
this.layers[i] = copyNetwork.layers[i]; //kopijuojamas kiekvienas sluoksnis kaip naujas
}
rnd = new Random(); //random rodiklis
InitNeurons(); //aprasomi neuronai
InitWeights(); //aprasomi svoriai
CopyWeights(copyNetwork.weights); //svoriu reiksmes perkopijuojamos is kito NN
} // NN()
private void Form1_Load(object sender, EventArgs e)
{
label2.Visible = false;
network = new NN();
Bitmap bmp = new Bitmap(168, 168);
pictureBox1.Image = bmp;
ifsPixelst = new FileStream(testpixelFile, FileMode.Open);
ifsLabelst = new FileStream(testlabelFile, FileMode.Open);
brImagest = new BinaryReader(ifsPixelst);
brLabelst = new BinaryReader(ifsLabelst);
for (int i = 0; i < 4; i++)
{
brImagest.ReadInt32();
}
brLabelst.ReadInt32();
brLabelst.ReadInt32();
}
static void Update(ref int generation, ref List <NN> nets, ref int populationSize, int[] layers)
{
SortNet(ref nets, populationSize); //surusiuoja didejancia tvarka
for (int i = 0; i < populationSize / 2; i++) //puse zemiausia arba auksciausi NN
{
nets[i] = new NN(nets[i + (populationSize / 2)]); //sukuria kopija tokia paties bet analogiskai geresnio NN
nets[i].Mutate();
nets[i + (populationSize / 2)] = new NN(nets[i + (populationSize / 2)]); //geriausi NN is naujo perdaromi identiskais
}
for (int i = 0; i < populationSize; i++) //kiekvienas NN
{
nets[i].SetFitness(0f); //nustato pagrinda kokybes lygiui
}
generation++; //generacija padideja
}
public static int LaunchSecond(NN net, int[] layers, ref string[] chosen, string reference)
{
StreamReader sk = new StreamReader(@"weightValues2.txt", System.Text.Encoding.GetEncoding(1257));
net.ReadWeights(sk);
sk.Close();
const int dataCount = 8; //kiek pradiniu duomenu
float[] primalData = new float[dataCount]; //naujas duomenu masyvas NN'ui
for (int i = 0; i < dataCount; i++)
{
primalData[i] = 1f;
}
primalData = TurnSecond(chosen, reference); //jau suzymetos klases ir langai paverciami i duomenis NN'ui
float output = ((net.FeedForward(primalData)[0]) + 1f) * 4f; //gaunamas outputas is neural network
Console.WriteLine("Second AI result: {0}", output);
List <float> possibilities = new List <float>(); //galimos vietos ikelti pamoka
for (int j = 0; j < dataCount; j++)
{
if (chosen[j] == "-")
{
possibilities.Add(j); //surenkamos visos tuscios vietos
}
}
float nearest = ClosestTo(possibilities, output); //surandama artimiausia tuscia vieta pamokai
return((int)nearest);
}
private void buttonReset_Click(object sender, EventArgs e)
{
nn = new NN(2, 1, 1, 4, 1);
pool = new GP(poolSize, nn.weightCount());
}
//Sekanti dalis skirta ikelti i button_press ar kazka tokio,
//cia todel kad negaliu tame darbe dabar keisti nieko, tai
//rasau cia
public static void GenerateAllTimetables(TTdata[] fullList, bool[][,] preferedTimes)
{
//pakeiciami duomenys i klases, reikalingas irasyti i
//lentele, t.y. klase arba kelios klases vienam mokytojui
//per viena pamoka. Po to pamokos surasomos mokytojams
const char symbol = '-';
List <string[, ]> allTimetables = new List <string[, ]>(); //pagrindine lentele
List <string> classList = new List <string>(); //ka irasyti vienam mokytojui
List <string> usedTeachers = new List <string>(); //jau surasyti mokytojai
List <string> classes = new List <string>(); //visos esamos klases
List <string> teacherList = new List <string>(); //visi mokytojai, gali ir kartotis
for (int i = 0; i < fullList.Length; i++)
{
teacherList.Add(fullList[i].GetTeacher());
//mokytoju sarasas (mokytojai kartojasi!)
//to reikia kad skirtingus dalykus vienam mokytojui rodytu skirtingose eilutese tvarkarastyje
//norint pakeisti, ideti if(teacherList.Contains(fullList[i].GetTeacher())) funkcija.
}
int[] firstLayers = new int[] { 40, 20, 10, 1 };
int[] secondLayers = new int[] { 8, 10, 10, 1 };
NN pickDay = new NN(firstLayers);
NN pickLesson = new NN(secondLayers);
for (int i = 0; i < fullList.Length; i++) //sarasas visu klasiu, be pasikartojimu
{
if (!classes.Contains(fullList[i].GetTeacher()))
{
classes.Add(fullList[i].GetClass());
}
}
for (int i = 0; i < classes.Capacity; i++) //visa tvarkarasti nustato i '-' simboli
{
for (int j = 0; j < teacherList.Capacity; j++)
{
for (int k = 0; k < 40; k++)
{
allTimetables[i][j, k] = symbol.ToString();
}
}
}
string[][,] complete = allTimetables.ToArray(); //sukuriamas specifinis trimatis masyvas
int[] timetableFitness = new int[classes.Capacity]; //sukuriamas masyvas stebeti tvarkarasciu kokybei
for (int u = 0; u < classes.Capacity; u++)
{
for (int i = 0; i < teacherList.Capacity; i++)
{
if (!usedTeachers.Contains(fullList[i].GetTeacher()))
{
classList = ExpandClass(fullList, fullList[i].GetTeacher(), preferedTimes[i]).ToList(); //sukuriamas sarasas mokytojo pamokoms
usedTeachers.Add(fullList[i].GetTeacher()); //kad to paties mokytojo neidetu i kelias klases
for (int j = 0; j < classList.Capacity; j++)
{
BeginFeeding(ref complete[u], i, classList[j], pickDay, pickLesson, firstLayers, secondLayers); //ikeliaos pamokos mokytojams po viena
}
}
}
timetableFitness[u] = GenerateClassTimetable(fullList, classes[u], complete[u], teacherList); //sudarytas vienas tvarkarastis
}
SortTimetables(ref complete, timetableFitness); //surusiuojami tvarkarasciai pagal ju kokybe, geriausi pirmieji
//Prideti isvedima i dataGridView
//Prideti esamo lango uzdaryma
//Prideti naujo lango (perziurejimo lenteles) atidaryma
}
public static void BeginFeeding(ref string[,] timetableMatrica, int teacherIndex, string toInput, /**/ NN firstNet, NN secondNet, int[] firstLayers, int[] secondLayers)
{
//init
//assumes timetableMatrica is not jagged array
string[] byName = new string[timetableMatrica.GetLength(0)];
float[] input;
string[] dayClasses = new string[8];
//set up data for feedforward
for (int i = 0; i < timetableMatrica.Length; i++)
{
for (int j = 0; j < timetableMatrica.GetLength(i); j++)
{
if (timetableMatrica[i, j] == toInput)
{
byName[i] = toInput;
}
else if (timetableMatrica[i, j] != " " && timetableMatrica[i, j] != "-" && timetableMatrica[i, j] != "")
{
byName[i] = timetableMatrica[i, j];
}
else
{
byName[i] = "-";
}
}
}
input = TurnFirst(byName, toInput);
float day = LaunchFirst(firstNet, firstLayers, input);
for (int i = 0; i < 8; i++)
{
dayClasses[i] = byName[5 * (int)day + i];
}
int lesson = LaunchSecond(secondNet, secondLayers, ref dayClasses, toInput);
timetableMatrica[teacherIndex, (int)day * 5 + lesson] = toInput;
}
static void Main(string[] args)
{
//Random r = new Random();
//while (true)
//{
// Console.WriteLine(r.NextDouble());
//}
Console.Title = "";
NN ntw = new NN(
layers_count: 1,
neurons_count: 3,
inputs_count: 2,
learning_rate: .1);
double[] vals = new double[4] {
0, 0, 0, 0
};
double[] oldvals = new double[4];// = vals;
vals[0] = ntw.GetResult(new double[] { 0, 0 });
Console.WriteLine(vals[0]);
vals[1] = ntw.GetResult(new double[] { 0, 1 });
Console.WriteLine(vals[1]);
vals[2] = ntw.GetResult(new double[] { 1, 0 });
Console.WriteLine(vals[2]);
vals[3] = ntw.GetResult(new double[] { 1, 1 });
Console.WriteLine(vals[3]);
oldvals[0] = vals[0];
oldvals[1] = vals[1];
oldvals[2] = vals[2];
oldvals[3] = vals[3];
//https://ai.stackexchange.com/questions/6167/what-is-the-best-xor-neural-network-configuration-out-there-in-terms-of-low-erro
//ntw.Network[1][1].Weights = new double[] { 10.0676, -6.6619, -6.3597 };
//ntw.Network[1][2].Weights = new double[] { 2.8261, -5.9874, -9.9025 };
//ntw.Network[2][0].Weights = new double[] { -4.6458, 9.461, -9.9307 };
for (int i = 0; i < 10000000; i++)
{
ntw.Teach(new double[] { 0, 1 }, 1);
ntw.Teach(new double[] { 1, 0 }, 1);
ntw.Teach(new double[] { 0, 0 }, 0);
ntw.Teach(new double[] { 1, 1 }, 0);
vals[0] = ntw.GetResult(new double[] { 0, 0 });
vals[1] = ntw.GetResult(new double[] { 0, 1 });
vals[2] = ntw.GetResult(new double[] { 1, 0 });
vals[3] = ntw.GetResult(new double[] { 1, 1 });
string title = "";
//title += (vals[0] == oldvals[0] ? "E" : vals[0] < oldvals[0] ? "D" : "U") + " ";
//title += (vals[1] == oldvals[1] ? "E" : vals[1] < oldvals[1] ? "D" : "U") + " ";
//title += (vals[2] == oldvals[2] ? "E" : vals[2] < oldvals[2] ? "D" : "U") + " ";
//title += (vals[3] == oldvals[3] ? "E" : vals[3] < oldvals[3] ? "D" : "U") + " ";
title += (vals[0] == oldvals[0] ? "-" : vals[0] < oldvals[0] ? "↓" : "↑") + " ";
title += (vals[1] == oldvals[1] ? "-" : vals[1] < oldvals[1] ? "↓" : "↑") + " ";
title += (vals[2] == oldvals[2] ? "-" : vals[2] < oldvals[2] ? "↓" : "↑") + " ";
title += (vals[3] == oldvals[3] ? "-" : vals[3] < oldvals[3] ? "↓" : "↑") + " ";
title += " ";
title += Math.Round(vals[0], 3) + " ";
title += Math.Round(vals[1], 3) + " ";
title += Math.Round(vals[2], 3) + " ";
title += Math.Round(vals[3], 3) + " ";
Console.Title = title;
oldvals[0] = vals[0];
oldvals[1] = vals[1];
oldvals[2] = vals[2];
oldvals[3] = vals[3];
Console.Write(i);
Console.Write(" ");
Console.Write(vals[0]);
Console.Write(" ");
Console.Write(vals[1]);
Console.Write(" ");
Console.Write(vals[2]);
Console.Write(" ");
Console.Write(vals[3]);
Console.WriteLine("");
//Console.Write(" ");
//Console.Write(ntw.GetResult(new double[] { 0, 0 }));
//Console.Write(" ");
//Console.Write(ntw.GetResult(new double[] { 0, 1 }));
//Console.Write(" ");
//Console.Write(ntw.GetResult(new double[] { 1, 0 }));
//Console.Write(" ");
//Console.Write(ntw.GetResult(new double[] { 1, 1 }));
//Console.WriteLine("");
}
//Console.WriteLine(ntw.GetResult(new double[] { 0.54, 0.2 }));
//Console.WriteLine(ntw.GetResult(new double[] { 0, 1 }));
//Console.WriteLine(ntw.GetResult(new double[] { 1, 0 }));
//Console.WriteLine(ntw.GetResult(new double[] { 1, 1 }));
//for (int i = 0; i < 10000; i++)
//{
// ntw.Teach(new double[] { 0, 0 }, 0);
// ntw.Teach(new double[] { 0, 1 }, 1);
// ntw.Teach(new double[] { 1, 0 }, 1);
// ntw.Teach(new double[] { 1, 1 }, 0);
//}
//Console.WriteLine("--");
//Console.WriteLine(ntw.GetResult(new double[] { 0, 0 }));
//Console.WriteLine(ntw.GetResult(new double[] { 0, 1 }));
//Console.WriteLine(ntw.GetResult(new double[] { 1, 0 }));
//Console.WriteLine(ntw.GetResult(new double[] { 1, 1 }));
Console.ReadLine();
}
