__info2_neural_network_file_format__

matrix.c

@@ -4,32 +4,156 @@
 
 // TODO Matrix-Funktionen implementieren
 
+//Matrix dimensionieren
 Matrix createMatrix(unsigned int rows, unsigned int cols)
 {
-    
+    Matrix m; //Struktur anlegen, Varibale m von Typ Matrix
+
+    //Sonderfall aus Unit-Test, wenn rows == 0 oder cols == 0, darf kein Speicher allokiert werden
+    if(rows==0 || cols==0){
+        m.rows = 0;
+        m.cols = 0;
+        m.buffer = NULL;
+        return m;
+}
+
+    //Normalfall
+    m.rows = rows;    // strukurvariable.feldname --> Struktur-Zugriffsoperator
+    m.cols = cols;
+    m.buffer = malloc((rows * cols) * sizeof(MatrixType)); //Speicher reserviert für Elemente
+
+    return m;
 }
 
 void clearMatrix(Matrix *matrix)
 {
-    
+    // falls Speicher existiert (buffer NICHT NULL ist): freigeben
+    if(matrix->buffer != NULL)
+    {
+        free(matrix->buffer);// Speicher freigegeben aber zeigt irgendwo hin (dangling pointer)
+    }
+
+    //Matrix in definierten leeren Zustand setzen
+    matrix->buffer = NULL; // dangling pointer zurücksetzen
+    matrix->rows = 0;
+    matrix->cols = 0;
 }
 
+//Wert in Matrix schreiben und wo genau
 void setMatrixAt(MatrixType value, Matrix matrix, unsigned int rowIdx, unsigned int colIdx)
 {
-    
+    // Sicherheit: wenn buffer == NULL (kein gültiger Speicher) oder Index außerhalb der Matrix --> return
+    if(matrix.buffer == NULL || rowIdx >= matrix.rows || colIdx >= matrix.cols)
+    {
+        return;
+    }
+    //index = Zeile * Anzahl_Spalten + Spalte
+    unsigned int index = rowIdx * matrix.cols + colIdx;
+
+    // Schreibt value direkt an die berechnete Position im Matrixspeicher
+    matrix.buffer[index] = value;
 }
 
 MatrixType getMatrixAt(const Matrix matrix, unsigned int rowIdx, unsigned int colIdx)
 {
-    
+    if (matrix.buffer == NULL || rowIdx >= matrix.rows || colIdx >= matrix.cols)
+    {
+        return 0;
+    }
+
+    unsigned int index = rowIdx * matrix.cols + colIdx;
+
+    return matrix.buffer[index];
 }
 
 Matrix add(const Matrix matrix1, const Matrix matrix2)
 {
-    
+    Matrix result;
+
+    //Zeilen müssen gleich sein
+    if (matrix1.rows != matrix2.rows)
+    {
+        result.rows = 0;
+        result.cols = 0;
+        result.buffer = NULL;
+        return result;
+    }
+
+
+    //Spalten müssen gleich sein (mit broadcasting)
+    //Fälle: gleiche Spalten ok, matrix1 hat 1 Spalte, matrix2 hat 1 Spalte
+    //sonst inkompatibel
+    if (matrix1.cols != matrix2.cols && matrix1.cols != 1 && matrix2.cols != 1)
+    {
+        result.rows = 0;
+        result.cols = 0;
+        result.buffer = NULL;
+        return result;
+    }
+
+    result.rows = matrix1.rows;
+    result.cols = (matrix1.cols > matrix2.cols) ? matrix1.cols : matrix2.cols;
+
+    result.buffer = malloc(result.rows * result.cols * sizeof(MatrixType));
+
+    for (unsigned int r = 0; r < result.rows; r++)
+    {
+        for (unsigned int c = 0; c < result.cols; c++)
+        {
+            // Bestimme Spalte für matrix1:
+            // Wenn nur 1 Spalte -> immer Spalte 0 benutzen
+            unsigned int c1 = (matrix1.cols == 1) ? 0 : c;
+
+            // Bestimme Spalte für matrix2:
+            unsigned int c2 = (matrix2.cols == 1) ? 0 : c;
+
+            MatrixType v1 = getMatrixAt(matrix1, r, c1);
+            MatrixType v2 = getMatrixAt(matrix2, r, c2);
+
+            setMatrixAt(v1 + v2, result, r, c);
+        }
+    }
+
+    return result;
 }
 
+
 Matrix multiply(const Matrix matrix1, const Matrix matrix2)
 {
-    
+    Matrix result;
+
+    if(matrix1.cols != matrix2.rows)
+    {
+    result.rows = 0;
+    result.cols = 0;
+    result.buffer = NULL;
+    return result;
+    }
+
+
+    result.rows = matrix1.rows;
+    result.cols = matrix2.cols;
+
+    result.buffer = malloc(result.rows * result.cols * sizeof(MatrixType));
+
+    for (unsigned int r = 0; r < result.rows; r++)
+    {
+        for (unsigned int c = 0; c < result.cols; c++)
+        {
+            MatrixType sum = 0;
+
+            // gemeinsame Dimension = matrix1.cols = matrix2.rows
+            for (unsigned int i = 0; i < matrix1.cols; i++)
+            {
+                MatrixType a = getMatrixAt(matrix1, r, i);
+                MatrixType b = getMatrixAt(matrix2, i, c);
+
+                sum += a * b;
+            }
+
+            setMatrixAt(sum, result, r, c);
+        }
+    }
+
+    return result;
 }

matrix.h

@@ -5,6 +5,13 @@
 
 typedef float MatrixType;
 
+// TODO Matrixtyp definieren
+typedef struct {
+    unsigned int rows;
+    unsigned int cols;
+    MatrixType* buffer; //buffer Pointer zeigt auf Heap, mit malloc dort dann Speicher reservieren
+} Matrix;
+
 
 Matrix createMatrix(unsigned int rows, unsigned int cols);
 void clearMatrix(Matrix *matrix);
@@ -13,4 +20,5 @@ MatrixType getMatrixAt(const Matrix matrix, unsigned int rowIdx, unsigned int co
 Matrix add(const Matrix matrix1, const Matrix matrix2);
 Matrix multiply(const Matrix matrix1, const Matrix matrix2);
 
+
 #endif

neuralNetworkTests.c

@@ -2,30 +2,23 @@
 #include <stdlib.h>
 #include <string.h>
 #include <math.h>
-#include "unity.h"
+#include "unity/unity.h"
 #include "neuralNetwork.h"
 
 
 static void prepareNeuralNetworkFile(const char *path, const NeuralNetwork nn)
 {
-
+        //öffnet die Datei in Binär zum schreiben
         FILE *file = fopen(path, "wb");
         if (!file) return;
-
+        //Fester Headerstring
         const char *tag = "__info2_neural_network_file_format__";
         fwrite(tag, sizeof(char), strlen(tag), file);
-
-        if (nn.numberOfLayers == 0 || nn.layers == NULL) {
-            int zero = 0;
-            fwrite(&zero, sizeof(int), 1, file); // inputDim
-            fwrite(&zero, sizeof(int), 1, file); // outputDim
-            fclose(file);
-            return;
-        }
-
+        //Bestimmt die Eingabedimension
         int inputDim = nn.layers[0].weights.cols;
         fwrite(&inputDim, sizeof(int), 1, file);
 
+
         for (unsigned int i = 0; i < nn.numberOfLayers; i++) {
 
             const Layer *L = &nn.layers[i];
@@ -33,12 +26,11 @@ static void prepareNeuralNetworkFile(const char *path, const NeuralNetwork nn)
             const Matrix *B = &L->biases;
 
             int outputDim = W->rows;
-
             fwrite(&outputDim, sizeof(int), 1, file);
-
+            //Anzahl der Weight-Werte
             size_t weightCount = (size_t)(W->rows * W->cols);
-            fwrite(W->buffer, sizeof(MatrixType), weightCount, file);
-
+            fwrite(&weightCount, sizeof(size_t), 1, file);
+            //Anzahl der Bias-Werte
             size_t biasCount = (size_t)(B->rows * B->cols);
             fwrite(B->buffer, sizeof(MatrixType), biasCount, file);