def test_feature_sampling(self):
num = 1024
data = generate_array_floats(n=num)
feat1 = Feature(data, random_seed=0)
feat2 = Feature(data, random_seed=0)
self.assertTrue(
numpy.array_equal(feat1.input_sample, feat2.input_sample))
def test_dataset_feature_names(self):
num = SAMPLE_SIZE_SMALL
feat1 = Feature(generate_array_floats(n=num), name="a")
feat2 = Feature(generate_array_floats(n=num), name="b")
dataset = DataSet([feat1, feat2])
self.assertEqual(num, dataset.count)
self.assertEqual(num, dataset.count)
self.assertTrue(
numpy.array_equal(feat1.values, dataset.features["a"].values))
self.assertTrue(
numpy.array_equal(feat2.values, dataset.features["b"].values))
self.assertTrue(numpy.array_equal(dataset["a", :num], feat1[:num]))
self.assertTrue(numpy.array_equal(dataset["b", :num], feat2[:num]))
def test_dataset_custom_loader(self):
num = SAMPLE_SIZE_SMALL
arr = generate_array_floats(n=num)
class MyCustomDataLoader(object):
def __len__(self):
return len(arr)
def __getitem__(self, idx):
return arr[idx]
dataset1 = DataSet([Feature(arr)], random_seed=0)
dataset2 = DataSet([Feature(MyCustomDataLoader())], random_seed=0)
self.assertTrue(
numpy.array_equal(dataset1.input_fn(), dataset2.input_fn()))
def test_feature_indexing(self):
num = 1024
data = generate_array_floats(n=num)
feat = Feature(data)
self.assertEqual(10, len(feat[:10]))
self.assertEqual(10, len(feat[0:10]))
self.assertEqual(10, len(feat[0:10:1]))
self.assertEqual(5, len(feat[0:10:2]))
def test_feature_init(self):
num = 1024
data = generate_array_floats(n=num)
feat = Feature(data)
self.assertEqual(num, len(feat))
self.assertEqual(num, len(feat[:]))
self.assertTrue(numpy.array_equal(data, feat.values))
self.assertTrue(numpy.array_equal(data[:num], feat[:num]))
self.assertNotEqual(feat.sample_var, -1)
def test_dataset_different_shapes(self):
num = SAMPLE_SIZE_SMALL
feat1 = Feature(generate_array_floats(n=num), name="feat1")
feat2 = Feature(generate_onehot_matrix(n=num), name="feat2")
dataset = DataSet([feat1, feat2])
self.assertEqual(num, dataset.count)
self.assertEqual(num, dataset.count)
self.assertTrue(
numpy.array_equal(feat1.values, dataset.features["feat1"].values))
self.assertTrue(
numpy.array_equal(feat2.values, dataset.features["feat2"].values))
self.assertTrue(numpy.array_equal(dataset["feat1", :num], feat1[:num]))
self.assertTrue(numpy.array_equal(dataset["feat2", :num], feat2[:num]))
arr1 = dataset[:, :num]
arr2 = [feat.values[:num] for feat in (feat1, feat2)]
for col1, col2 in zip(arr1, arr2):
self.assertTrue(numpy.array_equal(col1, col2))
def test_feature_custom_loader(self):
num = 1024
data = generate_array_floats(n=num)
class MyCustomDataLoader(object):
def __len__(self):
return len(data)
def __getitem__(self, idx):
return data[idx]
feat1 = Feature(data, random_seed=0)
feat2 = Feature(MyCustomDataLoader(), random_seed=0)
self.assertTrue(
numpy.array_equal(feat1.input_sample, feat2.input_sample))
self.assertTrue(numpy.array_equal(feat1.sampler(), feat2.sampler()))
def test_dataset_mismatch_len(self):
num = SAMPLE_SIZE_SMALL
feat1 = Feature(generate_array_floats(n=num))
feat2 = Feature(generate_array_floats(n=num * 2))
self.assertRaises(AssertionError, lambda: DataSet([feat1, feat2]))