Faulty Loss calculation in nn_scratch_end.py

[tribp]

I think there is a fault + inconsistency in how the Loss is calculated:
self.L_train -> is only calculate for 1 point in the train set -> should be the average for all training points
self.L_test -> is calculated as the sum of losses in the test set, but NOT divided by the nr of test samples.

Proposed solution:

class NeuralNetworkFromScratch:

       ...............

  def train(self, ITERATIONS=100):

        ...........

         ```
        # calc training loss
          L = np.sum(np.square(y_train_pred - y_train_true)).     ->>> remove
          self.L_train.append(L)    ->>> remove
        ```

        # Calculate the loss on the train set
        L_sum = 0
        for i in range(len(self.X_train)):
            y_true = self.y_train[i]
            y_pred = self.forward_pass(self.X_train[i])
            L_sum += np.sum((y_pred - y_true)**2)
        
        Nr_train_samples = len(self.X_train)
        self.Loss_train.append(L_sum/Nr_train_samples)

        # Calculate the loss on the test set
        L_sum = 0
        for i in range(len(self.X_test)):
            y_true = self.y_test[i]
            y_pred = self.forward_pass(self.X_test[i])
            L_sum += np.sum((y_pred - y_true)**2)
        
        Nr_test_samples = len(self.X_test)
        self.Loss_test.append(L_sum/Nr_test_samples)