def calc_fitness(pop): # the pop parameter is a list of gene combinations
rsts = list() # rsts will contain the generated individuals from all gene combinations
if len(pop[0]) > 20: # smaller k in k-fold CV for lower time cost when the length of gene combination is above 20
k = K - 2
else:
k = K
for gene_rg in pop:
indiv = Individual() # create a GA Individual for each gene combination in the pop parameter
indiv.ft_rg = gene_rg
indiv.gene_len = len(gene_rg)
mt_clf = np.array([0.0] * len(metrics_names))
X_cur = X[:, gene_rg]
# k fold cross validation
skf = StratifiedKFold(n_splits=k, random_state=random.randint(0, 100), shuffle=True)
clf = []
for train_index, test_index in skf.split(X_cur, y):
clf = GaussianNB()
X_train = X_cur[train_index, :]
X_test = X_cur[test_index, :]
y_train = y[train_index]
y_test = y[test_index]
clf.fit(X_train, y_train)
predicts = clf.predict(X_test)
result = confusion_matrix(y_test, predicts, labels=[0, 1])
mt_clf += np.array(calc_metrics(result))
indiv.metrics = list(map(lambda x: x / k, mt_clf))
indiv.clf = clf
indiv.acc = indiv.metrics[2]
if is_imbalanced: # deal with imbalanced data-sets
indiv.index = (0.6 * indiv.metrics[2] + 0.4 * indiv.metrics[3])
else:
indiv.index = indiv.acc
rsts.append(indiv)
return rsts
