working mnist dataloader

Author: kemsig

include/TensorSANN/data/Dataset.hpp

@@ -3,7 +3,7 @@
 template<typename T>
 class Dataset {
 public:
-    virtual size_t size() const = 0;
-    virtual std::pair<std::vector<T>, int> get_item(size_t index) const = 0;
+    virtual int size() const = 0;
+    // virtual std::pair<T, int> get_item(int index) const = 0;
     virtual ~Dataset() = default;
 };

include/TensorSANN/data/MNISTDataset.hpp

@@ -7,15 +7,15 @@
 
 namespace TensorSANN{
 
-class MNISTDataset : public Dataset<float>{
+class MNISTDataset : public Dataset<Tensor>{
     protected:
         std::vector<Tensor> images;
         std::vector<unsigned int> labels;
 
     public:
         MNISTDataset(std::string labels_path, std::string images_path);
-        size_t size() const override;
-        std::pair<Tensor, int> get_item(size_t index) const override;
+        int size() const override;
+        // std::pair<Tensor, int> get_item(int index) const override;
         void print_mnist_index(int index);
 
 

src/data/MNISTDataset.cpp

@@ -68,11 +68,11 @@ static std::vector<Tensor> load_mnist_images(std::string path){
 
     std::vector<Tensor> images;
     for (int i = 0; i < num_images; i++) {
-        std::vector<float> buffer(rows * cols);
+        std::vector<uint8_t> buffer(rows * cols);
         file.read(reinterpret_cast<char*>(buffer.data()), rows * cols);
-        
+        std::vector<float> float_buffer(buffer.begin(), buffer.end());
         //@@ temporary
-        images.push_back(Tensor({rows * cols, 1}, buffer) / 255.0);
+        images.push_back(Tensor({rows * cols, 1}, float_buffer) / 255.0);
     }
     file.close();
     return images;
@@ -85,12 +85,12 @@ MNISTDataset::MNISTDataset(std::string labels_path, std::string images_path){
 }
 
 
-size_t MNISTDataset::size(){
-    return 0;
-}
-std::pair<Tensor, int> MNISTDataset::get_item(size_t index){
-
+int MNISTDataset::size() const{
+    return 101;
 }
+// std::pair<Tensor, int> MNISTDataset::get_item(int index) const{
+//     return nullptr;
+// }
 
 static char pixel_to_char(float pixel_value) {
     if (pixel_value > 0.8f) return '#'; // Dark pixel
@@ -100,16 +100,17 @@ static char pixel_to_char(float pixel_value) {
     else return ' ';  // Very light pixel (almost white)
 }
 
-void TensorSANN::print_mnist_index(int index){
-    auto image = images[index].data;
+void MNISTDataset::print_mnist_index(int index){
+    auto image = images[index].data();
     int label = labels[index];
     std::cout << "Index: " << index << " Label: " << label << std::endl;
 
     for (int i = 0; i < MINIST_IMG_DIM; i++) {
         for (int j = 0; j < MINIST_IMG_DIM; j++) {
-            printf("%c ", pixel_to_char(image[i * MINIST_IMG_DIM + j]));
+            std::cout << pixel_to_char(image[i * MINIST_IMG_DIM + j]);
+            // std::cout << (iemage[i * MINIST_IMG_DIM + j]);
         }
-        printf("\n");
+        std::cout<< std::endl;
     }
 
 }

src/model_main.cpp

@@ -1,6 +1,7 @@
 #include <iostream>
 #include <vector>
 #include <memory>
+#include <string>
 
 #include "TensorSANN/utils/Tensor.hpp"
 #include "TensorSANN/layers/DenseLayer.hpp"
@@ -10,9 +11,16 @@
 #include "TensorSANN/activations/ReLU.hpp"
 #include "TensorSANN/optimizers/SGD.hpp"
 
+#include "TensorSANN/data/MNISTDataset.hpp"
+
 int main() {
     std::cout << "MODEL MAIN CPP ==========" << std::endl;
-    
+
+    std::string train_label_path = "../lib/datasets/mnist/t10k-labels.idx1-ubyte";
+    std::string train_images_path = "../lib/datasets/mnist/t10k-images.idx3-ubyte";
+    TensorSANN::MNISTDataset md(train_label_path, train_images_path);
+
+    md.print_mnist_index(0);
 
     std::vector<size_t> shape2 = {16,1};  // 2x3 tensor
     std::vector<float> data = {
@@ -43,7 +51,7 @@ int main() {
     layers.push_back(std::make_shared<TensorSANN::Softmax>());
 
 
-    int epoch = 2000;
+    int epoch = 1;
     for (int i = 1; i <= epoch; ++i){
         TensorSANN::Tensor fwd_op = input_tensor.transpose();
         // fwd