print("Execution time to fit 50 epochs: %s minutes" % ((end_time[1] - start_time[1])/60))

Execution time to fit 50 epochs: 8.164830807844798 minutes

nn_plot(hist[1], 50, loss=True, merge=False)

nn_plot(hist[1], 50, acc=True)

By evaluating the metrics we can see that this problem is still very simple to this model, it is still reaching overfitting. This is a good sign, it means that convolutionary neural networks are capable of solving more complex problems.

Classification: full digits set

We will proceed and apply the same model structure to the entire training set, with all 10 digits.

# One hot encoding
y_train_label = tf.keras.utils.to_categorical(y_train)
y_test_label = tf.keras.utils.to_categorical(y_test)

model.append(tf.keras.models.Sequential([
    tf.keras.layers.Conv2D(16, (3, 3), input_shape=(28, 28, 1)),
    tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),
    tf.keras.layers.Conv2D(32, (3, 3)),
    tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),
    tf.keras.layers.Conv2D(32, (3, 3)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(32, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
]))

params.append([tf.keras.losses.CategoricalCrossentropy(),
               tf.keras.optimizers.Adam(),
               tf.keras.metrics.CategoricalAccuracy()])

nn_compile(model[-1], params[-1])