def test_purge_and_embargo(self):
"""
Test that purging and embargo works correctly
"""
cv_gen = CombinatorialPurgedKFold(n_splits=6, n_test_splits=2, samples_info_sets=self.info_sets,
pct_embargo=0.05)
train_splits = []
test_splits = []
for train, test in cv_gen.split(self.info_sets):
train_splits.append(train)
test_splits.append(test)
cv_gen_no_embargo = CombinatorialPurgedKFold(n_splits=6, n_test_splits=2, samples_info_sets=self.info_sets,
pct_embargo=0.0)
train_splits_no = []
test_splits_no = []
for train, test in cv_gen_no_embargo.split(self.info_sets):
train_splits_no.append(train)
test_splits_no.append(test)
self.assertEqual(len(train_splits_no[0]) - len(train_splits[0]), 5) # Embargo of 1 train set
self.assertEqual(len(train_splits_no[1]) - len(train_splits[1]), 10) # Embargo of 2 train set
self.assertEqual(len(train_splits_no[-1]) - len(train_splits[-1]), 0) # Latest set is not embargo-ed
def test_test_times(self):
"""
Tests the Combinatorial CV with various number of train, test splits
"""
cv_gen = CombinatorialPurgedKFold(n_splits=6, n_test_splits=2, samples_info_sets=self.info_sets)
counter = 0 # number of splits counter
train_splits = [] # Array of train indices splits
test_splits = [] # Array of test indices splits
for train, test in cv_gen.split(self.info_sets):
counter += 1
train_splits.append(train)
test_splits.append(test)
self.assertEqual(counter, 15) # Example from the book says that for CombinatorialCV(6, 2) number of splits=15
# Check indices from example from the book (page 164)
for test_idx, [start_idx, end_idx] in zip(test_splits[0:5], [[0, 33], [0, 50], [0, 67], [0, 83], [0, 99]]):
self.assertEqual(test_idx[0], start_idx)
self.assertEqual(test_idx[-1], end_idx)
for test_idx, [start_idx, end_idx] in zip(test_splits[-5:], [[34, 83], [34, 99], [51, 83], [51, 99], [68, 99]]):
self.assertEqual(test_idx[0], start_idx)
self.assertEqual(test_idx[-1], end_idx)
# Test backtest paths
self.assertEqual(len(cv_gen.backtest_paths), cv_gen.n_splits - 1) # phi[N, 2] = N - 1
self.assertTrue((cv_gen.backtest_paths[0][0]['train'] == train_splits[0]).all())
self.assertTrue((cv_gen.backtest_paths[0][1]['train'] == train_splits[0]).all())
self.assertTrue((cv_gen.backtest_paths[1][0]['train'] == train_splits[1]).all())
self.assertTrue((cv_gen.backtest_paths[1][2]['train'] == train_splits[5]).all())
self.assertTrue((cv_gen.backtest_paths[-1][-1]['train'] == train_splits[-1]).all())
self.assertTrue((cv_gen.backtest_paths[-1][-2]['train'] == train_splits[-1]).all())
def test_errors_raise(self):
"""
Test if ValueErrors are raised
"""
with self.assertRaises(ValueError):
CombinatorialPurgedKFold(n_splits=6, n_test_splits=3, samples_info_sets=[1, 2, 3, 4])
cv_gen = CombinatorialPurgedKFold(n_splits=6, n_test_splits=2, samples_info_sets=self.info_sets,
pct_embargo=0.05)
with self.assertRaises(ValueError):
for _, _ in cv_gen.split(self.info_sets.iloc[2:]):
pass