def build_dnn_model()

in courses/DSL/challenge-mlprep/fraud_detection/trainer/model.py [0:0]

• 15 lines of code
• 4 McCabe index (conditional complexity)

def build_dnn_model(ds, num_bins, hash_bkts):
    
    inputs, transformed = get_input_and_transform(ds, num_bins, hash_bkts)
    
    # Concatenate preprocessed features into a single layer of the right dimension
    dnn_inputs = Concatenate()(transformed.values())
    dnn_inputs = Flatten()(dnn_inputs)

    # Create a DNN with 3 layers and ReLU activation. Output is predicted probability that a 
    # transaction is fraudulent.
    hid_1 = tf.keras.layers.Dense(64, activation='relu')(dnn_inputs)
    hid_2 = tf.keras.layers.Dense(32, activation='relu')(hid_1)
    hid_3 = tf.keras.layers.Dense(16, activation='relu')(hid_2)
    logit = tf.keras.layers.Dense(1, activation='sigmoid')(hid_3)  

    # List of metrics to be computed at training and evaluation time. 
    metrics = [tf.keras.metrics.BinaryAccuracy(),
               tf.keras.metrics.Precision(),
               tf.keras.metrics.Recall(),
               tf.keras.metrics.AUC(curve='PR')]

    # Create and compile the model with BinaryCrossentropy loss function
    model = tf.keras.Model(inputs=inputs, outputs=logit)
    model.compile(loss=tf.keras.losses.BinaryCrossentropy(from_logits=False), metrics=metrics)

    return model