keras_tuner_nLayers.py

#********************************************************************************
# we are using the conditional_scope of the keras_tuner hyperparameter class
#
# link: https://keras.io/api/keras_tuner/hyperparameters/#hyperparameters-class
#
# example by Gaston Mazzei, https://gastonmazzei.github.io/
#******************************************************************************** 


def MyHyperModel(HyperModel):
    """
    Build a Keras model that can tune 
    the # of hidden layers using "model_type" 
    as a variable that indicates the depth of 
    the model, i.e. 1,2,3... hidden layers

    (toy version, with the activation fixed at "relu" 
     and the number of neurons fixed at "32")
    """

    def build(self, hp, sh, depth, nclasses):

        # Instantiate a sequential model with "sh" as 
        # input shape
        model = Sequential()
        model.add(Input(shape=sh))

        # Set the possible network's depth as choices
        model_type = hp.Choice("model_type", [str(i) for i in range(1,depth+1)])
        for i in range(1,depth+1):
                if model_type == str(i):
                with hp.conditional_scope("model_type", [str(i)]):
                    for _ in range(i):
                        model.add(Dense(32, activation='relu')) # 32 neurons and "ReLu" for illustration purposes

        # Add the final classification layer and return the model
        model.add(Dense(nclasses, activation='softmax'))
        return model