def fixture_train_test_data(request, dataset) -> tuple:
"""Return a tuple of (x_train, x_test, y_train, y_test) in the
closed-world setting.
"""
sizes, times, classes = dataset
assert len(np.unique(classes)) == 3
if request.param == "sizes":
main_features = ensure_non_ragged(sizes)[:, :5000]
elif request.param == "timestamps":
main_features = extract_interarrival_times(times)[:, :5000]
else:
raise ValueError(f"Unknown param {request.param}")
assert main_features.shape[1] == 5000
metadata = (Metadata.COUNT_METADATA | Metadata.TIME_METADATA
| Metadata.SIZE_METADATA)
metadata_features = extract_metadata(
sizes=sizes, timestamps=times, metadata=metadata)
assert metadata_features.shape[1] == 12
features = np.hstack((main_features, metadata_features))
return train_test_split(
features, classes, stratify=classes, random_state=7152217)
def extract_features( # pylint: disable=unused-argument
self, sizes: np.ndarray, timestamps: np.ndarray) -> np.ndarray:
"""Extract size features for training and testing the classifier.
Override in a subclass to change the features used.
"""
return ensure_non_ragged(sizes, dimension=self.n_features_hint)
def test_ensure_non_ragged_crop_no_copy(ragged_data):
"""It should reduce the data to the specified dimension if no copy is
necessary.
"""
_, expected = ragged_data
result = ensure_non_ragged(expected, dimension=3)
np.testing.assert_array_equal(result, expected[:, :3])
def test_ensure_non_ragged(ragged_data):
"""Ensures that a ragged array is made not ragged.
"""
data, expected = ragged_data
result = ensure_non_ragged(data, copy=False)
np.testing.assert_array_equal(result, expected)
assert not np.shares_memory(result, expected)
def fixture_train_test_sizes(dataset) -> tuple:
"""Return a tuple of (x_train, x_test, y_train, y_test) in the
open-world setting.
"""
sizes, _, classes = dataset
features = ensure_non_ragged(sizes)[:, :5000]
assert len(np.unique(classes)) == 3
assert features.shape[1] == 5000
return train_test_split(
features, classes, stratify=classes, random_state=7141845)
def extract_features(self, sizes: np.ndarray,
timestamps: np.ndarray) -> np.ndarray:
metadata = (Metadata.COUNT_METADATA | Metadata.TIME_METADATA
| Metadata.SIZE_METADATA)
meta_features = extract_metadata(sizes=sizes,
timestamps=timestamps,
metadata=metadata,
batch_size=5_000)
if self.n_meta_features != meta_features.shape[1]:
raise ValueError(
f"The number of metadata features, {meta_features.shape[1]}, "
f"does not match the amount specified: {self.n_meta_features}."
)
if self.feature_type == "sizes":
features = ensure_non_ragged(sizes, dimension=self.n_features_hint)
else:
assert self.feature_type == "time"
features = extract_interarrival_times(
timestamps, dimension=self.n_features_hint)
return np.hstack((features, meta_features))
def test_ensure_non_ragged_pad(ragged_data):
"""It should reduce the data to the specified dimension."""
data, expected = ragged_data
result = ensure_non_ragged(data, dimension=6)
np.testing.assert_array_equal(result, np.pad(expected, [(0, 0), (0, 2)]))
def test_ensure_non_ragged_crop(ragged_data):
"""It should reduce the data to the specified dimension."""
data, expected = ragged_data
result = ensure_non_ragged(data, dimension=3)
np.testing.assert_array_equal(result, expected[:, :3])
def test_ensure_non_ragged_copy(ragged_data):
"""Should copy when copy=True"""
_, expected = ragged_data
result = ensure_non_ragged(expected, copy=True)
np.testing.assert_array_equal(result, expected)
assert not np.shares_memory(result, expected)
def test_ensure_non_ragged_noop(ragged_data):
"""If already not ragged, it should not be changed"""
_, expected = ragged_data
result = ensure_non_ragged(expected, copy=False)
np.testing.assert_array_equal(result, expected)
assert np.shares_memory(result, expected)