Inital Commit

NeuronalNetworkLib/NeuronalNetworkLib.sln

@@ -0,0 +1,25 @@
+
+Microsoft Visual Studio Solution File, Format Version 12.00
+# Visual Studio Version 16
+VisualStudioVersion = 16.0.29806.167
+MinimumVisualStudioVersion = 10.0.40219.1
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "NeuronalNetworkLib", "NeuronalNetworkLib\NeuronalNetworkLib.csproj", "{0F576159-6F91-4902-8F29-B9C3C0477A39}"
+EndProject
+Global
+	GlobalSection(SolutionConfigurationPlatforms) = preSolution
+		Debug|Any CPU = Debug|Any CPU
+		Release|Any CPU = Release|Any CPU
+	EndGlobalSection
+	GlobalSection(ProjectConfigurationPlatforms) = postSolution
+		{0F576159-6F91-4902-8F29-B9C3C0477A39}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{0F576159-6F91-4902-8F29-B9C3C0477A39}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{0F576159-6F91-4902-8F29-B9C3C0477A39}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{0F576159-6F91-4902-8F29-B9C3C0477A39}.Release|Any CPU.Build.0 = Release|Any CPU
+	EndGlobalSection
+	GlobalSection(SolutionProperties) = preSolution
+		HideSolutionNode = FALSE
+	EndGlobalSection
+	GlobalSection(ExtensibilityGlobals) = postSolution
+		SolutionGuid = {14E13FD8-414F-4623-B3C3-01EA7D096C22}
+	EndGlobalSection
+EndGlobal

NeuronalNetworkLib/NeuronalNetworkLib/ActivisionFunctions/Booleanactivation.cs

@@ -0,0 +1,24 @@
+using NeuronalNetworkLib.ActivisionFunctions.Interface;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.ActivisionFunctions
+{
+    /// <summary>
+    /// Booleanaktivierung alles grösser 0 = 1
+    /// </summary>
+    public class Booleanactivation : IActivationFunction
+    {
+        public float Activation(float input)
+        {
+            if (input <= 0) return 0;
+            return 1;
+        }
+
+        public float Derivation(float input)
+        {
+            return 1;
+        }
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/ActivisionFunctions/HyperbolicTanActivation.cs

@@ -0,0 +1,28 @@
+using NeuronalNetworkLib.ActivisionFunctions.Interface;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.ActivisionFunctions
+{
+    /// <summary>
+    /// Pendelt werte zwischen -1 und 1 ein
+    /// </summary>
+    public class HyperbolicTanActivation : IActivationFunction
+    {
+        public float Activation(float input)
+        {
+            float Epx = MathF.Pow(MathF.E, input); 
+            float Enx = MathF.Pow(MathF.E, -input);
+
+            return (Epx - Enx) / (Epx + Enx);
+        }
+
+        public float Derivation(float input)
+        {
+            float epx = 4 * (MathF.Pow(MathF.E, 2 * input));
+            float enx = MathF.Pow((MathF.Pow(MathF.E, 2 * input) + 1), 2);
+            return epx / enx;
+        }
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/ActivisionFunctions/IdentityActivation.cs

@@ -0,0 +1,23 @@
+using NeuronalNetworkLib.ActivisionFunctions.Interface;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.ActivisionFunctions
+{
+    /// <summary>
+    /// Ausgabe des Input Wertes
+    /// </summary>
+    public class IdentityActivation : IActivationFunction
+    {
+        public float Activation(float input)
+        {
+            return input;
+        }
+
+        public float Derivation(float input)
+        {
+            return 1;
+        }
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/ActivisionFunctions/Interface/IActivationFunction.cs

@@ -0,0 +1,51 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.ActivisionFunctions.Interface
+{   
+/// <summary>
+/// Interface für Neuronaktivierungsfunktionen
+/// </summary>
+    public interface IActivationFunction
+    {
+        /// <summary>
+        /// Booleanaktivierung alles grösser 0 = 1
+        /// </summary>
+        public static Booleanactivation booleanactivation = new Booleanactivation();
+
+        /// <summary>
+        /// Pendelt werte zwischen -1 und 1 ein
+        /// </summary>
+        public static HyperbolicTanActivation hyperbolicTanActivation = new HyperbolicTanActivation();
+
+        /// <summary>
+        /// Ausgabe des Input Wertes
+        /// </summary>
+        public static IdentityActivation identityActivation = new IdentityActivation();
+
+        /// <summary>
+        /// Giebt input wert aus wenn görsser Null ansonsten wird -0.1 ausgegeben (eigentlich null allerdings 0.1 besser für lernefeckt)
+        /// </summary>
+        public static ReLuActivation reLuActivation = new ReLuActivation();
+
+        /// <summary>
+        /// Pendelt input wert zwischen 0 und 1 ein
+        /// </summary>
+        public static SigmoidActivation sigmoidActivation = new SigmoidActivation();
+
+        /// <summary>
+        /// Aktivierungsfunktion Aktivieren
+        /// </summary>
+        /// <param name="input">Summe aller NeuronInputs</param>
+        /// <returns>Wert des Neurons nach Aktivierungsfunktion</returns>
+        public float Activation(float input);
+
+        /// <summary>
+        /// Differential der Aktivierungsfunktion für Backpropagation
+        /// </summary>
+        /// <param name="input"></param>
+        /// <returns></returns>
+        public float Derivation(float input);
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/ActivisionFunctions/ReLuActivation.cs

@@ -0,0 +1,38 @@
+using NeuronalNetworkLib.ActivisionFunctions.Interface;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.ActivisionFunctions
+{
+    /// <summary>
+    /// Giebt input wert aus wenn görsser Null ansonsten wird -0.1 ausgegeben (eigentlich null allerdings 0.1 besser für lernefeckt)
+    /// </summary>
+    public class ReLuActivation : IActivationFunction
+    {
+
+        public float Activation(float input)
+        {
+            if (input >= 0)
+            {
+                return input;
+            }
+            else
+            {
+                return -0.1f;
+            }
+        }
+
+        public float Derivation(float input)
+        {
+            if (input >= 0)
+            {
+                return 1;
+            }
+            else
+            {
+                return 0;
+            }
+        }
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/ActivisionFunctions/SigmoidActivation.cs

@@ -0,0 +1,24 @@
+using NeuronalNetworkLib.ActivisionFunctions.Interface;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.ActivisionFunctions
+{
+    /// <summary>
+    /// Pendelt input wert zwischen 0 und 1 ein
+    /// </summary>
+    public class SigmoidActivation : IActivationFunction
+    {
+        public float Activation(float input)
+        {
+            return 1 / (1 + MathF.Pow(MathF.E, -input));
+        }
+        public float Derivation(float input)
+        {
+            float epx = MathF.Pow(MathF.E, -input);
+            float enx = MathF.Pow((1 + MathF.Pow(MathF.E, -input)), 2);
+            return epx / enx;
+        }
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Class1.cs

@@ -0,0 +1,8 @@
+using System;
+
+namespace NeuronalNetworkLib
+{
+    public class NeuronalNetwork
+    {
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Connection/Abstract/ConnectionAbstract.cs

@@ -0,0 +1,15 @@
+using NeuronalNetworkLib.Neurons;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.Connection
+{
+    public abstract class ConnectionAbstract
+    {
+        public abstract Neuron getNeuron();
+        public abstract float getWeight();
+        public abstract float getValue();
+        public abstract void addWeight(float weightdelta);
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Connection/ConnectionBatch.cs

@@ -0,0 +1,52 @@
+using NeuronalNetworkLib.Neurons;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.Connection
+{
+    public class ConnectionBatch : ConnectionAbstract
+    {
+        private Neuron _neuron;
+        private float _weight;
+        private float momentum = 0;
+        private float weightadd;
+
+        public ConnectionBatch(Neuron neuron, float weight)
+        {
+            _neuron = neuron;
+            _weight = weight;
+        }
+
+        public override float getValue()
+        {
+            return _neuron.getvalue() * _weight;
+        }
+
+        public override void addWeight(float weightdelta)
+        {
+            weightadd += weightdelta;
+
+        }
+
+        public void applyBatch()
+        {
+            momentum += weightadd;
+            momentum *= 0.9f;
+            _weight += weightadd + momentum;
+
+            weightadd = 0;
+        }
+
+        public override Neuron getNeuron()
+        {
+            return _neuron;
+        }
+
+        public override float getWeight()
+        {
+            return _weight;
+        }
+
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Connection/ConnectionNormal.cs

@@ -0,0 +1,42 @@
+using NeuronalNetworkLib.Neurons;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.Connection
+{
+    public class ConnectionNormal : ConnectionAbstract
+    {
+        private Neuron _neuron;
+        private float _weight;
+        private float momentum = 0;
+
+        public ConnectionNormal(Neuron neuron, float weight)
+        {
+            _neuron = neuron;
+            _weight = weight;
+        }
+
+        public override Neuron getNeuron()
+        {
+            return _neuron;
+        }
+
+        public override float getValue()
+        {
+            return _neuron.getvalue() * _weight;
+        }
+
+        public override float getWeight()
+        {
+            return _weight;
+        }
+
+        public override void addWeight(float weightdelta)
+        {
+            momentum += weightdelta;
+            momentum *= 0.9f;
+            _weight += weightdelta + momentum;
+        }
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/NeuronalNetworkLib.csproj

@@ -0,0 +1,13 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+  <PropertyGroup>
+    <TargetFramework>netstandard2.1</TargetFramework>
+  </PropertyGroup>
+
+  <ItemGroup>
+    <Folder Include="LoadParameters\" />
+    <Folder Include="Learnfunctions\" />
+    <Folder Include="SaveParameters\" />
+  </ItemGroup>
+
+</Project>

NeuronalNetworkLib/NeuronalNetworkLib/NeuronalNetworks/BatchNetwork.cs

@@ -0,0 +1,10 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.NeuronalNetworks
+{
+    class BatchNetwork
+    {
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/NeuronalNetworks/BatchlessNetwork.cs

@@ -0,0 +1,116 @@
+using NeuronalNetworkLib.ActivisionFunctions.Interface;
+using NeuronalNetworkLib.Connection;
+using NeuronalNetworkLib.Neurons;
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.NeuronalNetworks
+{
+    class BatchlessNetwork
+    {
+        private List<InputNeuron> inputNeurons = new List<InputNeuron>();
+        private List<WorkingNeuron> outputneurons = new List<WorkingNeuron>();
+        private List<WorkingNeuron> hiddenneurons = new List<WorkingNeuron>();
+        private List<int> hiddenlayers = new List<int>();
+        Random random = new Random();
+
+
+        public WorkingNeuron CreateNewOutput()
+        {
+            WorkingNeuron output = new WorkingNeuron();
+            outputneurons.Add(output);
+            return output;
+        }
+
+        public InputNeuron CreateNewInput()
+        {
+            InputNeuron input = new InputNeuron();
+            inputNeurons.Add(input);
+            return input;
+        }
+
+        public void CreateHiddenLayer(int amount, IActivationFunction activationfunction)
+        {
+            for(int i= 0; i< amount; i++)
+            {
+                WorkingNeuron wn = new WorkingNeuron();
+                wn.SetActiavtionFunction(activationfunction);
+                hiddenneurons.Add(wn);
+            }
+            hiddenlayers.Add(amount);
+        }
+
+        public void Reset()
+        {
+            foreach(WorkingNeuron wn in outputneurons)
+            {
+                wn.Reset();
+            }
+            foreach(WorkingNeuron wn in hiddenneurons)
+            {
+                wn.Reset();
+            }
+        }
+
+        public void CreateFullMesh()
+        {
+            if(hiddenneurons.Count == 0)
+            {
+               foreach(WorkingNeuron wn in outputneurons)
+                {
+                    foreach(InputNeuron inp in inputNeurons)
+                    {
+                        float weight = (float)random.NextDouble() * 10; 
+                        wn.Addconnection(new ConnectionNormal(inp,weight));
+                    }
+                }
+            }
+            else
+            {
+                int counter = 0;
+                bool first = false;
+                for(int j = 0; j < hiddenlayers.Count; j++)
+                {
+                    if (!first)
+                    {
+                        for(int i = 0; i < hiddenlayers[j]; i++)
+                        {
+                            foreach(InputNeuron inp in inputNeurons)
+                            {
+                                float weight = (float)random.NextDouble() * 10;
+                                hiddenneurons[i].Addconnection(new ConnectionNormal(inp, weight)); 
+                            }
+                        }
+                        counter += hiddenlayers[j];
+                        first = true;
+                    }
+                    else
+                    {
+                        for(int i = counter; i < counter + hiddenlayers[j]; i++)
+                        {
+                            for(int h = counter - hiddenlayers[j - 1]; h < counter; h++)
+                            {
+                                float weight = (float)random.NextDouble() * 10;
+                                hiddenneurons[i].Addconnection(new ConnectionNormal(hiddenneurons[h],weight));
+                            }
+                        }
+                        counter += hiddenlayers[j];
+                    }
+                }
+
+                foreach(WorkingNeuron wn in outputneurons)
+                {
+                    for(int i = hiddenlayers[hiddenlayers.Count-1]; i < hiddenlayers.Count; i++)
+                    {
+                        float weight = (float)random.NextDouble() * 10;
+                        wn.Addconnection(new ConnectionNormal(hiddenneurons[i], weight));
+                    }
+                }
+            }
+        }
+
+
+
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/NeuronalNetworks/WorkingNetwork.cs

@@ -0,0 +1,10 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.NeuronalNetworks
+{
+    class WorkingNetwork
+    {
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Neurons/Abstract/Neuron.cs

@@ -0,0 +1,11 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.Neurons
+{
+    public abstract class Neuron
+    {
+        public abstract float getvalue();
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Neurons/BiasNeuron.cs

@@ -0,0 +1,23 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.Neurons
+{
+    public class BiasNeuron : Neuron
+    {
+
+        private float _Value;
+
+        public override float getvalue()
+        {
+            return _Value;
+        }
+
+        public void setValue(float value)
+        {
+            _Value = value;
+        }
+
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Neurons/InputNeuron.cs

@@ -0,0 +1,21 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace NeuronalNetworkLib.Neurons
+{
+    public class InputNeuron : Neuron
+    {
+        private float _Value;
+
+        public override float getvalue()
+        {
+            return _Value;
+        }
+
+        public void setValue(float value)
+        {
+            _Value = value;
+        }
+    }
+}

NeuronalNetworkLib/NeuronalNetworkLib/Neurons/WorkingNeuron.cs

@@ -0,0 +1,141 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+using NeuronalNetworkLib.ActivisionFunctions.Interface;
+using NeuronalNetworkLib.Connection;
+
+namespace NeuronalNetworkLib.Neurons
+{
+    public class WorkingNeuron : Neuron
+    {
+        private List<ConnectionAbstract> connections = new List<ConnectionAbstract>();
+        private IActivationFunction activationfunction = IActivationFunction.identityActivation;
+        private float _smalldelta = 0;
+        private float _value = 0;
+        private bool _valueClean = false;
+
+        /// <summary>
+        /// Berechnet ausgabe wert mittels aller input und der Aktivierungsfunktion
+        /// </summary>
+        /// <returns>Ausgabewert des Neurons</returns>
+        public override float getvalue()
+        {
+            if (!_valueClean)
+            {
+                float sum = 0;
+                foreach(ConnectionAbstract c in connections)
+                {
+                    sum += c.getValue();
+                }
+                _value = activationfunction.Activation(sum);
+                _valueClean = true;
+            }
+            return _value;
+        }
+
+        /// <summary>
+        /// Giebt die Summe aller Inputs zurück wichtig für Learning
+        /// </summary>
+        /// <returns>Summe aller Inputs</returns>
+        public float Getsum()
+        {
+            float sum = 0;
+
+            foreach (ConnectionAbstract c in connections)
+            {
+                sum += c.getValue();
+            }
+
+            return sum;
+        }
+
+        /// <summary>
+        /// Neue Connection zum Workingneuron hinzufügen
+        /// </summary>
+        /// <param name="c">Connection die hinzugefügt werden soll</param>
+        public void Addconnection(ConnectionAbstract c)
+        {
+            connections.Add(c);
+        }
+
+        /// <summary>
+        /// Berechnung der Outputabweichung
+        /// </summary>
+        /// <param name="should">Soll Output</param>
+        public void CalculateOutputDelta(float should)
+        {
+            _smalldelta = should - getvalue();
+        }
+
+        /// <summary>
+        /// Rücksetzen des neurons für neuen lernvorgang
+        /// </summary>
+        public void Reset()
+        {
+            _smalldelta = 0;
+            _valueClean = false;
+        }
+
+        /// <summary>
+        /// Aktivierungsfunktion für neuron festlegen
+        /// </summary>
+        /// <param name="activationfunction">Activationfunktion für das Neuron</param>
+        public void SetActiavtionFunction(IActivationFunction activationfunction)
+        {
+            this.activationfunction = activationfunction;
+        }
+
+        /// <summary>
+        /// Smaldelta an hintere Ebene zurückgeben für messabweichungsberechnung
+        /// </summary>
+        public void BackpropagateSmallDelta()
+        {
+            foreach(ConnectionAbstract c in connections)
+            {
+                Neuron n = c.getNeuron();
+                if (n is WorkingNeuron)
+                {
+                    WorkingNeuron wn = (WorkingNeuron)n;
+                    wn._smalldelta += _smalldelta * c.getWeight();
+                }
+            }
+        }
+
+        /// <summary>
+        /// Deltalernrgel mit Backpropagation leraning 
+        /// Berechnung(Lernstärke(epsilon) * (Auswertungsabweichung(smalldelta) * ableitung Actfunktion(sumedereingänge)) * ausgang des Connectetn neurons)
+        /// </summary>
+        /// <param name="epsilon">Lernstärke</param>
+        public void deltaleraning(float epsilon)
+        {
+            _smalldelta = _smalldelta * activationfunction.Derivation(Getsum());
+
+            for (int i = 0; i < connections.Count; i++)
+            {
+                float bigdelta = epsilon * _smalldelta * connections[i].getNeuron().getvalue();
+
+                connections[i].addWeight(bigdelta);
+            }
+        }
+
+        /// <summary>
+        /// Betch activierung der Connections für Batchleraning Netzwerke
+        /// </summary>
+        public void ApplyBatch()
+        {
+            foreach(ConnectionAbstract c in connections)
+            {
+                if (c is ConnectionBatch)
+                {
+                    ConnectionBatch cb = (ConnectionBatch)c;
+
+                    cb.applyBatch();
+                }
+                else
+                {
+                    throw new Exception("Batchlearning is activated");
+                }
+            }
+        }
+    }
+}