VincentGat

def get_data():
    """Synthetic binary classification dataset."""
    data, targets = make_classification(
        n_samples=50000,
        n_features=45,
        n_informative=12,
        n_redundant=7,
        random_state=134985745,
    )
    return data, targets

VincentGat

from deap.algorithms import eaSimple
a = eaSimple(pop, toolbox, cxpb=0.3, mutpb=0.1, ngen=50 verbose=True)

VincentGat

toolbox.register("select", tools.selTournament, tournsize=2)

VincentGat

def mutate(indiv, indpb):
    for param_name, param_value in indiv.params.items():
        if random.random() < indpb:
            indiv.params[param_name] = indiv.generate_hyperparams()[param_name]


    indiv.features = np.array([feat if random.random() > indpb else not feat for feat in indiv.features])
    return indiv,

toolbox.register("mutate", mutate, indpb=0.05)

VincentGat

from deap import base, creator, tools

creator.create("FitnessMax", base.Fitness, weights=(1.0,))
creator.create("Individual", Model, fitness=creator.FitnessMax)

toolbox = base.Toolbox()
toolbox.register("population", tools.initRepeat, list, creator.Individual)
toolbox.register("evaluate", evaluate)

pop = toolbox.population(100)

VincentGat

def evaluate(indiv):
    rkf = RepeatedStratifiedKFold(n_splits=10, n_repeats=3, random_state=SEED)
    data = x_train[:, indiv.features]
    perfo = []
    for train_index, test_index in rkf.split(X=data, y=y_train):
        x_train_split, x_test_split = data[train_index], data[test_index]
        y_train_split, y_test_split = y_train[train_index], y_train[test_index]
        model = indiv.model
        model.fit(x_train_split, y_train_split)
        y_pred = model.predict(x_test_split)

VincentGat

# repair rf if broken
if self.params['max_features'] > np.sum(self.features):
    self.params['max_features'] = random.randint(1, max(2, np.sum(self.features)))

VincentGat

class Model:
    nb_features = x.shape[1]

    def __init__(self):
        self.model_family = random.choice(['nb', 'rf', 'lgbm', 'catboost'], p=[0, 0.3334, 0.3333, 0.3333])
        self._features = random.choice([True, False], self.nb_features)
        self.params = self.generate_hyperparams()

    @property
    def features(self):

VincentGat

model_family = random.choice(['nb', 'rf', 'lgbm', 'catboost'], p=[0, 0.3334, 0.3333, 0.3333])
features = random.choice([True, False], nb_features)

VincentGat

def cxModel(ind1, ind2, swap_indpb=0.5):
    new_feat_ind1 = []
    new_feat_ind2 = []
    for feature1, feature2 in zip(ind1.features, ind2.features):
        if random.random() < swap_indpb:
            new_feat_ind1.append(feature2)
            new_feat_ind2.append(feature1)
        else:
            new_feat_ind1.append(feature1)
            new_feat_ind2.append(feature2)