In the following section, we will use Keras. Keras is a Python library for neural networks and provides a high-level interface to TensorFlow libraries. We do not intend to give a complete tutorial on Keras or CNN, but we want to show how we can use Matplotlib to visualize the loss function, accuracy, and outliers of the results. 

Readers who are not familiar with machine learning should be able to go through the logic of the remaining chapter and hopefully understand why visualizing the loss function, accuracy, and outliers of the results is important in fine-tuning the CNN model. 

Here is a snippet of code for the CNN; the most important part is the evaluation section after this!

# Import sklearn models for preprocessing input data
from sklearn.model_selection import train_test_split 
from sklearn.preprocessing import LabelBinarizer

# Import the necessary Keras libraries
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers.convolutional import Convolution2D, MaxPooling2D
from keras import backend as K
from keras.callbacks import History 

# Randomize and split data into training dataset with right format to feed to Keras
lb = LabelBinarizer()
X = np.expand_dims(digits.images.T, axis=0).T
y = lb.fit_transform(digits.target)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=100)

# Start a Keras sequential model
model = Sequential()

# Set input format shape as (batch, height, width, channels)
K.set_image_data_format('channels_last') # inputs with shape (batch, height, width, channels)

model.add(Convolution2D(filters=4,kernel_size=(3,3),padding='same',input_shape=(8,8,1),activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2)))

# Drop out 5% of training data in each batch
model.add(Flatten())
model.add(Dropout(0.05))
model.add(Dense(10, activation= 'softmax'))

# Set variable 'history' to store callbacks to track the validation loss
history = History()

# Compile the model
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

# Fit the model and save the callbacks of validation loss and accuracy to 'history'
model.fit(X_train,y_train, epochs=100, batch_size= 128, callbacks=[history])