def single_split(data, estimator, scoring):
attrs, classes = utils.horizontal_split(data)
X_train, X_test, y_train, y_test = train_test_split(attrs,
classes,
test_size=0.4)
estimator.fit(X_train, y_train)
scorer = get_scorer(scoring)
return scorer(estimator, X_test, y_test)
def load(file):
iterator = csv.reader(open(file, "r"))
data = list(iterator)
class_index = utils.get_class_index(data)
for i in range(len(data)):
for j in range(len(data[i])):
val = data[i][j]
data[i][j] = float(val) if j != class_index else val
return utils.horizontal_split(data)
def get_attr_by_class_probs(self, X, y):
data = utils.merge_attrs(X, y)
class_index = utils.get_class_index(data)
result = self.get_empty_classes_dict([])
for record in data:
class_key = record[class_index]
result[class_key].append(record)
for key in result:
class_X, class_y = utils.horizontal_split(result[key])
result[key] = self.get_attr_probs(class_X, self.empty_bins())
return result
def score(data,
estimator_const,
est_param,
k=10,
cross_val=True,
stratified=True):
attrs, classes = utils.horizontal_split(data)
attr_ranges = utils.attr_ranges(attrs)
unique_classes = utils.unique_classes(
classes
) # it's important to recognize classes from both training and test set(whole data)
estimator = estimator_const(unique_classes, est_param, attr_ranges)
scoring = ['f1_macro', 'accuracy', 'precision_macro', 'recall_macro']
score = validation.k_fold(data, estimator, scoring, k, stratified) \
if cross_val else validation.single_split(data, estimator, scoring)
print(("mean " if cross_val else "") + f"score: {score}")
return score
def k_fold(data, estimator, scoring, k, stratified=True):
attrs, classes = utils.horizontal_split(data)
cv = StratifiedKFold(n_splits=k) if stratified else KFold(n_splits=k)
scores = cross_validate(estimator, attrs, classes, cv=cv, scoring=scoring)
scores_test = [scores["test_" + scoring_elem] for scoring_elem in scoring]
return np.mean(scores_test, (0, 1)) # mean by metrics then by folds