Backup Save

include/neural_network/model/model.hpp

@@ -323,7 +323,7 @@ struct model{
      *  
      *
      */
-    bool add_optimizer_sgd(const panic::types::real_t learning_rate = static_cast<panic::types::real_t>(1e-3));
+    bool add_optimizer_sgd(const panic::types::real_t learning_rate = static_cast<panic::types::real_t>(1));
 
 
 
@@ -337,6 +337,22 @@ struct model{
      */
     bool finalize();
 
+
+    /**
+     * @brief Optimizes the model with parameters on trainable layers
+     *
+     * Computes:
+     * @code
+     * model.optimize()
+     * @endcode
+     *
+     * @return true if optimization is done correctly.
+     *
+     */
+    bool optimize();
+
+
+
     /**
      * @brief Trains the model with input data
      *

include/neural_network/optimizers/optimizer.hpp

@@ -24,7 +24,7 @@
  * Module Name:       neural_network
  * File Name:         optimizer.hpp
  * Revision:          0.1.0
- * Date:              23-06-2026
+ * Date:              06-08-2026
  * Author:            Michelle Bausager
  *
  * Description:

include/neural_network/optimizers/optimizer_sgd.hpp

@@ -52,7 +52,6 @@ struct optimizer_sgd: optimizer{
 
      panic::types::real_t learning_rate;
 
-
     /**
      * @brief Constructor
      *

main.cpp

@@ -705,7 +705,7 @@ int main(void) {
     panic::tensor::real_matrix X;
     panic::tensor::uint_vector y;
 
-    panic::types::uint_t samples = 10;
+    panic::types::uint_t samples = 100;
     panic::types::uint_t classes = 3;
 
 
@@ -716,7 +716,7 @@ int main(void) {
     panic::neural_network::model mymodel;
 
     // Create Dense layer with 2 input features and 3 output values
-    if (!mymodel.add_layer_dense(2,3)){
+    if (!mymodel.add_layer_dense(2,64)){
         return false;
     }
     
@@ -726,7 +726,7 @@ int main(void) {
     }
     
     // Create a second dense layer with 3 inputs and 3 outputs
-    if (!mymodel.add_layer_dense(3, 3)){
+    if (!mymodel.add_layer_dense(64, 3)){
         return false;
     }
     
@@ -740,15 +740,19 @@ int main(void) {
     if (!mymodel.add_loss_categorical_crossentropy()){
         return false;
     }
+
+    if (! mymodel.add_optimizer_sgd()){
+        return false;
+    }
    
     if (!mymodel.finalize()){
         return false;
     }
 
-    //mymodel.add_optimizer_sgd();
     
-    panic::types::uint_t epochs = 10;
+    
-    panic::types::uint_t print_every = 1;
+    panic::types::uint_t epochs = 10000;
+    panic::types::uint_t print_every = 100;
 
     if (!mymodel.train(X, y, epochs, print_every)){
         std::cout << "Training failed" << std::endl;

src/neural_network/model/model.cpp

@@ -606,6 +606,38 @@ bool model::finalize(){
 }
 
 
+//--------------------------------------------------------------------------------------------------------------------------
+// Function Name : panic::neural_network::model::optimize
+//
+// Description:
+//      Optimizes the model with parameters on trainable layers
+//--------------------------------------------------------------------------------------------------------------------------
+bool model::optimize(){
+
+       if (optimizer_function == 0){
+        return false;
+    }
+
+
+    
+    for(panic::types::uint_t i = 0; i < trainable_layer_count; ++i){
+        
+        // Validate the stored layer pointers before using them.
+        if (trainable_layers[i] == 0){
+            return false;
+        }
+
+        if (! optimizer_function->update_params(*trainable_layers[i]) ){
+            return false;
+        }
+
+    }
+
+
+    return true;
+}
+
+
 
 //--------------------------------------------------------------------------------------------------------------------------
 // Function Name : panic::neural_network::model::train
@@ -668,7 +700,9 @@ bool model::train(const panic::tensor::real_matrix& X_train,
         }
 
 
-        //optimize();
+        if (! optimize()){
+            return false;
+        }
 
     }
 

src/neural_network/optimizers/optimizer_sgd.cpp

@@ -50,7 +50,7 @@
  * Small vectors and matrices are kept serial because the overhead of starting
  * worker threads can be larger than the work itself.
  */
-static const panic::types::uint_t optimizer_sgd_omp_min_size = 500;
+static const panic::types::uint_t optimizer_sgd_omp_min_size = 250;
 //---------------------------------------------------------------------------------------------------------------------------
 // INPLEMENTATION
 //---------------------------------------------------------------------------------------------------------------------------
@@ -77,21 +77,36 @@ optimizer_sgd::optimizer_sgd(const panic::types::real_t learning_rate) {
 //--------------------------------------------------------------------------------------------------------------------------
 bool optimizer_sgd::update_params(trainable_layer& layer) {
 
+
+    // Gradients must match their corresponding parameters.
+    if (layer.weights.rows() != layer.dweights.rows() || layer.weights.cols() != layer.dweights.cols() || layer.biases.size() != layer.dbiases.size()){
+        return false;
+    }
+
+    if (layer.weights.rows() == static_cast<panic::types::uint_t>(0) || layer.weights.cols() == static_cast<panic::types::uint_t>(0) || layer.biases.size() == static_cast<panic::types::uint_t>(0) ){
+        return false;
+    }
+
+
     panic::tensor::real_matrix weight_updates;
     panic::tensor::real_vector bias_updates;
 
-    if (!panic::math::mul(layer.weights, -learning_rate, weight_updates)){
+    // weight_updates = -learning_rate * dweights
+    if (!panic::math::mul(layer.dweights, -learning_rate, weight_updates)){
     	return false;
     }
 
+    // weights += weight_updates
     if (!panic::math::add(layer.weights, weight_updates, layer.weights)){
     	return false;
     }
 
-    if (!panic::math::mul(layer.biases, -learning_rate, bias_updates)){
+    // bias_updates = -learning_rate * dbiases
+    if (!panic::math::mul(layer.dbiases, -learning_rate, bias_updates)){
     	return false;
     }
 
+    // biases += bias_updates
     if (!panic::math::add(layer.biases, bias_updates, layer.biases)){
     	return false;
     }