Optimazation is done

include/neural_network/layer/trainable_layer.hpp

@@ -63,6 +63,24 @@ struct trainable_layer:layer{
     panic::tensor::real_matrix dweights;
     panic::tensor::real_vector dbiases;
 
+    /**
+     * @brief Previous parameter updates used by momentum SGD.
+     *
+     * These remain empty unless an optimizer using momentum
+     * initializes them.
+     */
+    panic::tensor::real_matrix weight_momentums;
+    panic::tensor::real_vector bias_momentums;
+
+    /**
+     * @brief Previous parameter updates used by AdaGrad.
+     *
+     * These remain empty unless an optimizer using cache
+     * initializes them.
+     */
+    panic::tensor::real_matrix weight_cache;
+    panic::tensor::real_vector bias_cache;
+
 
 	virtual ~trainable_layer() = default;