[edit] Minor modifications

Signed-off-by: Simone Rossi <simone.rossi@eurecom.fr>

DL_lab1/NeuralNetwork.py

@@ -57,7 +57,7 @@ class NeuralNetwork(object):
         else:
             self.W_hidden_to_output = np.random.uniform(size = (self.hidden_layer_size, self.output_layer_size)) / np.sqrt(self.hidden_layer_size)
 
-    def train(self, data, validation_data, iterations=50, learning_rate=5.0, verbose=False, batch_size=1):
+    def train(self, data, validation_data, iterations=50, learning_rate=5.0, verbose=False, batch_size=1, plot=True):
         start_time = time.time()
         training_accuracies = []
         validation_accuracies = []
@@ -69,14 +69,15 @@ class NeuralNetwork(object):
         best_h2o_W = self.W_hidden_to_output
            # SIMONE
         for it in range(iterations):
-            error = [0]
+            error = []
             for batch_index in range(0,len(inputs), batch_size):   # SIMONE
                 inputs_batch = inputs[batch_index : min(batch_index+batch_size, len(inputs))]
                 targets_batch = targets[batch_index : min(batch_index+batch_size, len(targets))]
                 self.feedforward(inputs_batch)
                 self.backpropagate(targets_batch, learning_rate=learning_rate)
-                #error.append(list(targets - self.o_output))
-            error = np.array(error).flatten()
+                error.append(targets_batch - self.o_output)
+
+            error = np.concatenate(error).flatten()
             error *= error
             training_accuracies.append(100*self.predict(data)/len(data[0]))
             validation_accuracies.append(100*self.predict(validation_data)/len(validation_data[0]))
@@ -84,11 +85,14 @@ class NeuralNetwork(object):
                 best_i2h_W = self.W_input_to_hidden
                 best_h2o_W = self.W_hidden_to_output
             if verbose:
-                print("[Iteration %2d/%2d]  -Training_Accuracy:  %2.2f %%  -Validation_Accuracy: %2.2f %%  -time: %2.2f " %(it+1, iterations,
-                                                            training_accuracies[-1], validation_accuracies[-1], time.time() - start_time))
-                print("    - MSE:", np.sum(error)/len(targets))
-        print("Training time:", time.time()-start_time)
-        plot_train_val(range(1, iterations+1), training_accuracies, validation_accuracies, "Accuracy")
+                print("[Iteration %2d/%2d]  -Training_Accuracy:  %2.2f %%  -Validation_Accuracy: %6.2f %%  -time: %6.2f " %(it+1, iterations,
+                                                            training_accuracies[-1], validation_accuracies[-1], time.time() - start_time), end=' ')
+                print(" - MSE: %.5f" % (np.sum(error)/len(targets)))
+        print("Training time: %.4f s" % (time.time()-start_time))
+        print('Final accurary: %6.2f%%' % validation_accuracies[-1])
+        if plot:
+            plot_train_val(range(1, iterations+1), training_accuracies, validation_accuracies, "Accuracy")
+        return validation_accuracies
        
     def train_xe(self, data, validation_data, iterations=50, learning_rate=5.0, verbose=False):
         start_time = time.time()