def test_fold_epoch_vs_epoch(self):
df = pd.DataFrame({"a": np.arange(10)})
fe = [
d.x["a"].tolist()
for d in Sampler(XYWeight(df[["a"]], df[["a"]]),
fold_epochs=2).sample_for_training()
]
e = [
d.x["a"].tolist() for d in Sampler(XYWeight(df[["a"]], df[["a"]]),
epochs=2).sample_for_training()
]
self.assertListEqual(fe, e)
def test_filter(self):
sampler = Sampler(XYWeight(TEST_DF, TEST_DF),
splitter=lambda i, *args: (i[:-3], i[-3:]),
filter=lambda idx, y: y["Close"] > 300)
batches = list(sampler.sample_for_training())
self.assertEqual(1, len(batches))
self.assertEqual(101, len(batches[0].x))
self.assertEqual(101, len(batches[0].y))
def test_cross_validation_multiindex_row(self):
sampler = Sampler(XYWeight(TEST_MUTLI_INDEX_ROW_DF,
TEST_MUTLI_INDEX_ROW_DF),
splitter=None,
cross_validation=PartitionedOnRowMultiIndexCV(
KFold(2)),
epochs=1)
batches = list(sampler.sample_for_training())
self.assertEqual(4, len(batches))
def test_infinite_generator(self):
sampler = Sampler(XYWeight(TEST_DF, None, TEST_DF.tail()),
splitter=None,
epochs=None)
for i, s in enumerate(sampler.sample_for_training()):
if i > 100:
break
self.assertGreater(i, 100)
def test_multiframe_decorator(self):
sampler = Sampler(XYWeight(
MultiFrameDecorator((TEST_DF.tail(), TEST_DF)), TEST_DF.tail(10)),
splitter=lambda i, *args: (i[:3], i[3:]),
epochs=2)
batches = list(sampler.sample_for_training())
self.assertEqual(2, len(batches))
self.assertEqual(3, len(batches[0].x))
self.assertEqual(3, len(batches[1].y))
def test_simple_batches(self):
sampler = Sampler(XYWeight(TEST_DF, None, TEST_DF.tail()),
splitter=lambda i: (i[:3], i[3:]),
batch_size=2)
batches = list(sampler.sample_for_training())
self.assertEqual(2, len(batches))
self.assertEqual(2, len(batches[0].x))
self.assertEqual(
2, len(batches[1].x)
) # in case of one single instance add the 2nd last as well
def test_simple_sample_split(self):
def check_test(test_data):
self.assertEqual(1, len(test_data))
self.assertEqual(2, len(test_data[0].x))
self.assertIsNone(test_data[0].y)
sampler = Sampler(XYWeight(TEST_DF, None, TEST_DF.tail()),
splitter=lambda i, *args: (i[:3], i[3:]),
epochs=2,
after_fold=check_test)
batches = list(sampler.sample_for_training())
self.assertEqual(2, len(batches))
self.assertEqual(3, len(batches[0].weight))
self.assertIsNone(batches[0].y)
def test_simple_sample(self):
def check_test(test_data):
self.assertEqual(0, len(test_data))
sampler = Sampler(XYWeight(TEST_DF, None, TEST_DF.tail()),
splitter=None,
after_fold=check_test)
batches = list(sampler.sample_for_training())
self.assertEqual(1, len(batches))
batch1 = batches[0]
self.assertEqual(5, len(batch1.x))
self.assertEqual(5, len(batch1.weight))
self.assertIsNone(batch1.y)
self.assertIsNone(batch1.y)
def test_simple_sample_split_multiindex_row(self):
def check_test(test_data):
self.assertIn("A", test_data[0].x.index)
self.assertIn("B", test_data[0].x.index)
sampler = Sampler(XYWeight(TEST_MUTLI_INDEX_ROW_DF),
splitter=naive_splitter(
0.5, partition_row_multi_index=True),
after_fold=check_test,
epochs=1)
samples = list(sampler.sample_for_training())
self.assertEqual(1, len(samples))
self.assertEqual(4, len(samples[0].x))
self.assertIn("A", samples[0].x.index)
self.assertIn("B", samples[0].x.index)
def test_callback_early_stop(self):
class Stopper(object):
def __init__(self):
self.i = 0
def callback(self, epoch):
self.i += 1
if self.i > 2:
raise StopIteration
sampler = Sampler(XYWeight(TEST_MUTLI_INDEX_ROW_DF,
TEST_MUTLI_INDEX_ROW_DF),
splitter=None,
fold_epochs=100,
after_fold_epoch=Stopper().callback)
self.assertEqual(3, len(list(sampler.sample_for_training())))
def test_cross_validation(self):
def check_test(test_data):
self.assertEqual(1, len(test_data))
self.assertEqual(3, len(test_data[0]))
sampler = Sampler(XYWeight(TEST_DF, TEST_DF, TEST_DF.tail(30)),
splitter=lambda i, *args: (i[:-3], i[-3:]),
cross_validation=KFold(3).split,
after_fold=check_test,
epochs=1)
batches = list(sampler.sample_for_training())
self.assertEqual(3, len(batches))
batch1, batch2, batch3 = batches
self.assertEqual(18, len(batch1.x))
self.assertEqual(18, len(batch2.y))
self.assertEqual(18, len(batch3.weight))
def test_multiple_test_data(self):
def check_test(test_data):
self.assertEqual(3, len(test_data))
self.assertEqual('19930203',
test_data[0].x.index[0].strftime("%Y%m%d"))
self.assertEqual('19930202',
test_data[0].x.index[1].strftime("%Y%m%d"))
self.assertEqual('19930201',
test_data[-1].x.index[0].strftime("%Y%m%d"))
self.assertEqual('19930129',
test_data[-1].x.index[1].strftime("%Y%m%d"))
sampler = Sampler(XYWeight(TEST_DF),
splitter=lambda i, *args: (i[:4], i[-3:]),
cross_validation=lambda i:
[([0, 1, 2], np.array([[3, 2, 1], [2, 1, 0]]))],
after_fold=check_test)
batches = list(sampler.sample_for_training())
self.assertEqual(1, len(batches))
self.assertEqual(3, len(batches[0].x))