def main(): # pragma: no cover
# pylint: disable=import-outside-toplevel
from Orange.evaluation import ShuffleSplit
data = Table("brown-selected")
if not "test_rows": # change to `if not "test_rows" to test columns
data = append_column(data, "M", StringVariable("Test"),
(np.arange(len(data)).reshape(-1, 1) %
30).astype(str))
res = ShuffleSplit(n_resamples=5, test_size=0.7, stratified=False)
indices = iter(res.get_indices(data))
datasets = []
for i in range(1, 6):
sample, _ = next(indices)
data1 = data[sample]
data1.name = chr(ord("A") + i)
datasets.append((data1, i))
else:
domain = data.domain
data1 = data.transform(Domain(domain.attributes[:15],
domain.class_var))
data2 = data.transform(Domain(domain.attributes[10:],
domain.class_var))
datasets = [(data1, 1), (data2, 2)]
WidgetPreview(OWVennDiagram).run(setData=datasets)
def test_results(self):
data = self.random_table
train_size, n_resamples = 0.6, 10
res = ShuffleSplit(data, [NaiveBayesLearner()], train_size=train_size,
test_size=1 - train_size, n_resamples=n_resamples)
self.assertEqual(len(res.predicted[0]),
n_resamples * self.nrows * (1 - train_size))
def test_init(self):
res = ShuffleSplit(n_resamples=1, train_size=0.1, test_size=0.2,
stratified=False, random_state=42)
self.assertEqual(res.n_resamples, 1)
self.assertEqual(res.train_size, 0.1)
self.assertEqual(res.test_size, 0.2)
self.assertFalse(res.stratified)
self.assertEqual(res.random_state, 42)
def test_results(self):
nrows, ncols = 100, 10
data = random_data(nrows, ncols)
train_size, n_resamples = 0.6, 10
res = ShuffleSplit(data, [NaiveBayesLearner()], train_size=train_size,
test_size=1 - train_size, n_resamples=n_resamples)
self.assertEqual(len(res.predicted[0]),
n_resamples * nrows * (1 - train_size))
def test_stratified(self):
# strata size
n = 50
res = ShuffleSplit(
train_size=.5, test_size=.5, n_resamples=3, stratified=True,
random_state=0)(self.iris, [NaiveBayesLearner()])
for fold in res.folds:
self.assertEqual(np.count_nonzero(res.row_indices[fold] < n), n // 2)
self.assertEqual(np.count_nonzero(res.row_indices[fold] < 2 * n), n)
def test_not_stratified(self):
# strata size
n = 50
res = ShuffleSplit(self.iris, [NaiveBayesLearner()],
train_size=.5, test_size=.5,
n_resamples=3, stratified=False, random_state=0)
strata_samples = []
for train, test in res.indices:
strata_samples.append(np.count_nonzero(train < n) == n/2)
strata_samples.append(np.count_nonzero(train < 2 * n) == n)
self.assertTrue(not all(strata_samples))
def test_not_stratified(self):
# strata size
n = 50
res = ShuffleSplit(
train_size=.5, test_size=.5, n_resamples=3, stratified=False,
random_state=0)(self.iris, [NaiveBayesLearner()])
strata_samples = []
for fold in res.folds:
strata_samples += [
np.count_nonzero(res.row_indices[fold] < n) == n // 2,
np.count_nonzero(res.row_indices[fold] < 2 * n) == n]
self.assertTrue(not all(strata_samples))