INNER CODE UNIT · Python
run
joeddav/devol · devol/devol.py:88
def run(self, dataset, num_generations, pop_size, epochs, fitness=None,
metric='accuracy'):
"""
Run genetic search on dataset given number of generations and
population size
Args:
dataset : tuple or list of numpy arrays in form ((train_data,
train_labels), (validation_data, validation_labels))
num_generations (int): number of generations to search
pop_size (int): initial population size
epochs (int): epochs for each model eval, passed to keras model.fit
fitness (None, optional): scoring function to be applied to
population scores, will be called on a numpy array which is
a min/max scaled version of evaluated model metrics, so It
should accept a real number including 0. If left as default
just the min/max scaled values will be used.
metric (str, optional): must be "accuracy" or "loss" , defines what