Finish Conv2D and Flatten implementations

CMakeLists.txt

@@ -31,16 +31,19 @@ include_directories(
 
 # Add source files
 set(SOURCE_FILES
-    src/core/module.cpp
     src/core/tensor.cpp
+    src/utils/tensor_utils.cpp
+    src/core/module.cpp
     src/modules/layers/linear.cpp
+    src/modules/layers/conv2d.cpp
+    src/modules/layers/flatten.cpp
+    src/utils/conv2d_utils.cpp
     src/modules/layers/dropout.cpp
     src/modules/layers/conv2d.cpp
     src/modules/losses/mse.cpp
     src/modules/activations/relu.cpp
     src/modules/activations/softmax.cpp
     src/modules/losses/cross_entropy.cpp
-    src/utils/tensor_utils.cpp
     src/datasets/mnist.cpp
     src/models/mlp.cpp
     src/metrics/accuracy.cpp
@@ -51,7 +54,7 @@ set(SOURCE_FILES
 add_library(neuralnet ${SOURCE_FILES})
 
 # Add the executable for the main example
-add_executable(main examples/test_tensor.cpp)
+add_executable(main examples/test_conv2d.cpp)
 target_link_libraries(main neuralnet)
 
 # Only build tests if BUILD_TESTS is ON

debug.sh

@@ -0,0 +1,4 @@
+cd build/
+cmake -DCMAKE_BUILD_TYPE=Debug ..
+make
+lldb main

examples/main.cpp

@@ -6,7 +6,8 @@
 #include "utils.hpp"
 using namespace nn;
 
-int main() {
+int main()
+{
 
     // Define the hyperparameters
 
@@ -21,41 +22,48 @@ int main() {
     const string mnist_label_file = "../data/mnist/train-labels.idx1-ubyte";
 
     // load MNIST data
-    if (!dataset.load_data(mnist_image_file, mnist_label_file)) {
+    if (!dataset.load_data(mnist_image_file, mnist_label_file))
+    {
         cerr << "Failed to load dataset" << endl;
         return 1;
     }
 
     // Initialize the model
     MLP model = MLP({784, 128, 64, 10}, DROPOUT_P);
 
+    cout << "Finished model initialization" << endl;
+
     // Define the loss function
     CrossEntropyLoss criterion = CrossEntropyLoss();
 
+    cout << "Finished loss initialization" << endl;
+
     double loss = 0.0;
     double acc = 0.0;
     vector<double> loss_list;
     vector<double> accuracy_list;
 
+    cout << "Training started..." << endl;
+
     // // Train the model
     // Example of iterating through all batches
-    for (size_t e = 0; e < EPOCH; e++) {
+    for (size_t e = 0; e < EPOCH; e++)
+    {
         cout << "\nEpoch " << e + 1 << ":\n";
-        dataset.reset();  // Reset batch counter at the start of each epoch
+        dataset.reset(); // Reset batch counter at the start of each epoch
         loss_list.clear();
         accuracy_list.clear();
-
-        for (size_t i = 0; i < dataset.get_num_batches(); i++) {
+
+        for (size_t i = 0; i < dataset.get_num_batches(); i++)
+        {
             auto batch = dataset.get_next_batch();
             auto [data, labels] = batch.to_tensor();
 
             // forward propagation
             Tensor<> output = model(data);
 
             loss = criterion(output, labels);
-            // cout << "After loss" << endl;
             acc = metrics::accuracy(output, labels);
-            // cout << "After acc" << endl;
 
             accuracy_list.push_back(acc);
             loss_list.push_back(loss);
@@ -66,7 +74,7 @@ int main() {
             model.update_params(LR);
 
             // print the training stats
-            print_training_stats_line(i, loss, acc);
+            print_stats_line(i, loss, acc);
         }
 
         double total_loss = accumulate(loss_list.begin(), loss_list.end(), 0.0) / loss_list.size();
@@ -78,5 +86,51 @@ int main() {
         cout << "------------------------------------" << endl;
     }
 
+    // Inference
+
+    model.eval();
+
+    const string mnist_image_file_test = "../data/mnist/t10k-images.idx3-ubyte";
+    const string mnist_label_file_test = "../data/mnist/t10k-labels.idx1-ubyte";
+
+    MNIST test_dataset(BATCH_SIZE);
+
+    if (!test_dataset.load_data(mnist_image_file_test, mnist_label_file_test))
+    {
+        cerr << "Failed to load test dataset" << endl;
+        return 1;
+    }
+
+    cout << "\n------------------------------------" << endl;
+    cout << "Testing started..." << endl;
+
+    loss = 0.0;
+    acc = 0.0;
+    loss_list.clear();
+    accuracy_list.clear();
+
+    for (size_t i = 0; i < test_dataset.get_num_batches(); i++)
+    {
+        auto batch = test_dataset.get_next_batch();
+        auto [data, labels] = batch.to_tensor();
+
+        // forward propagation
+        Tensor<> output = model(data);
+
+        loss = criterion(output, labels);
+        acc = metrics::accuracy(output, labels);
+
+        accuracy_list.push_back(acc);
+        loss_list.push_back(loss);
+
+        // print the testing stats
+        print_stats_line(i, loss, acc);
+    }
+
+    double total_loss = accumulate(loss_list.begin(), loss_list.end(), 0.0) / loss_list.size();
+    double total_acc = accumulate(accuracy_list.begin(), accuracy_list.end(), 0.0) / accuracy_list.size() * 100;
+
+    cout << "------------------------------------" << endl;
+
     return 0;
 }