def setUp(self):
self.cwd = os.getcwd()
tests_dir = __file__
os.chdir(os.path.dirname(tests_dir))
decoder = arff.ArffDecoder()
with open(os.path.join("datasets", "dataset.arff")) as fh:
dataset = decoder.decode(fh, encode_nominal=True)
# -1 because the last attribute is the class
self.attribute_types = [
'numeric' if type(type_) != list else 'nominal'
for name, type_ in dataset['attributes'][:-1]]
self.categorical = [True if attribute == 'nominal' else False
for attribute in self.attribute_types]
data = np.array(dataset['data'], dtype=np.float64)
X = data[:,:-1]
y = data[:,-1].reshape((-1,))
ohe = OneHotEncoder(self.categorical)
X_transformed = ohe.fit_transform(X)
imp = Imputer(copy=False)
X_transformed = imp.fit_transform(X_transformed)
standard_scaler = StandardScaler()
X_transformed = standard_scaler.fit_transform(X_transformed)
X_transformed = X_transformed.todense()
# Transform the array which indicates the categorical metafeatures
number_numerical = np.sum(~np.array(self.categorical))
categorical_transformed = [True] * (X_transformed.shape[1] -
number_numerical) + \
[False] * number_numerical
self.categorical_transformed = categorical_transformed
self.X = X
self.X_transformed = X_transformed
self.y = y
self.mf = meta_features.metafeatures
self.helpers = meta_features.helper_functions
# Precompute some helper functions
self.helpers.set_value("PCA", self.helpers["PCA"]
(self.X_transformed, self.y))
self.helpers.set_value("MissingValues", self.helpers[
"MissingValues"](self.X, self.y, self.categorical))
self.helpers.set_value("NumSymbols", self.helpers["NumSymbols"](
self.X, self.y, self.categorical))
self.helpers.set_value("ClassOccurences",
self.helpers["ClassOccurences"](self.X, self.y))
self.helpers.set_value("Skewnesses",
self.helpers["Skewnesses"](self.X_transformed, self.y,
self.categorical_transformed))
self.helpers.set_value("Kurtosisses",
self.helpers["Kurtosisses"](self.X_transformed, self.y,
self.categorical_transformed))
def test_scaler_1d(self):
"""Test scaling of dataset along single axis"""
rng = np.random.RandomState(0)
X = rng.randn(5)
X_orig_copy = X.copy()
scaler = StandardScaler()
X_scaled = scaler.fit(X).transform(X, copy=False)
assert_array_almost_equal(X_scaled.mean(axis=0), 0.0)
assert_array_almost_equal(X_scaled.std(axis=0), 1.0)
# check inverse transform
X_scaled_back = scaler.inverse_transform(X_scaled)
assert_array_almost_equal(X_scaled_back, X_orig_copy)
# Test with 1D list
X = [0., 1., 2, 0.4, 1.]
scaler = StandardScaler()
X_scaled = scaler.fit(X).transform(X, copy=False)
assert_array_almost_equal(X_scaled.mean(axis=0), 0.0)
assert_array_almost_equal(X_scaled.std(axis=0), 1.0)
X_scaled = scale(X)
assert_array_almost_equal(X_scaled.mean(axis=0), 0.0)
assert_array_almost_equal(X_scaled.std(axis=0), 1.0)
# Test with sparse list
X = scipy.sparse.coo_matrix((np.random.random((10,)),
([i**2 for i in range(10)],
[0 for i in range(10)])))
X = X.tocsr()
scaler = StandardScaler()
X_scaled = scaler.fit(X).transform(X, copy=False)
self.assertFalse(np.any(np.isnan(X_scaled.data)))
self.assertAlmostEqual(X_scaled.mean(), 0)
self.assertAlmostEqual(np.sqrt(X_scaled.data.var()), 1)
# Check that X has not been copied
# self.assertTrue(X_scaled is X)
# Check that the matrix is still sparse
self.assertEqual(len(X.indices), 10)
def test_standard_scaler_sparse_boston_data(self):
X_train, Y_train, X_test, Y_test = get_dataset('boston',
make_sparse=True)
num_data_points = len(X_train.data)
scaler = StandardScaler()
scaler.fit(X_train, Y_train)
tr = scaler.transform(X_train)
# Test this for every single dimension!
means = np.array([tr.data[tr.indptr[i]:tr.indptr[i + 1]].mean()
for i in range(13)])
vars = np.array([tr.data[tr.indptr[i]:tr.indptr[i + 1]].var()
for i in range(13)])
for i in chain(range(1, 3), range(4, 13)):
self.assertAlmostEqual(means[i], 0, 2)
self.assertAlmostEqual(vars[i], 1, 2)
self.assertAlmostEqual(means[3], 1)
self.assertAlmostEqual(vars[3], 0)
# Test that the matrix is still sparse
self.assertTrue(scipy.sparse.issparse(tr))
self.assertEqual(num_data_points, len(tr.data))
def setUp(self):
self.cwd = os.getcwd()
tests_dir = __file__
os.chdir(os.path.dirname(tests_dir))
decoder = arff.ArffDecoder()
with open(os.path.join("datasets", "dataset.arff")) as fh:
dataset = decoder.decode(fh, encode_nominal=True)
# -1 because the last attribute is the class
self.attribute_types = [
'numeric' if type(type_) != list else 'nominal'
for name, type_ in dataset['attributes'][:-1]
]
self.categorical = [
True if attribute == 'nominal' else False
for attribute in self.attribute_types
]
data = np.array(dataset['data'], dtype=np.float64)
X = data[:, :-1]
y = data[:, -1].reshape((-1, ))
ohe = OneHotEncoder(self.categorical)
X_transformed = ohe.fit_transform(X)
imp = Imputer(copy=False)
X_transformed = imp.fit_transform(X_transformed)
standard_scaler = StandardScaler()
X_transformed = standard_scaler.fit_transform(X_transformed)
X_transformed = X_transformed.todense()
# Transform the array which indicates the categorical metafeatures
number_numerical = np.sum(~np.array(self.categorical))
categorical_transformed = [True] * (X_transformed.shape[1] -
number_numerical) + \
[False] * number_numerical
self.categorical_transformed = categorical_transformed
self.X = X
self.X_transformed = X_transformed
self.y = y
self.mf = meta_features.metafeatures
self.helpers = meta_features.helper_functions
# Precompute some helper functions
self.helpers.set_value("PCA", self.helpers["PCA"](self.X_transformed,
self.y))
self.helpers.set_value(
"MissingValues", self.helpers["MissingValues"](self.X, self.y,
self.categorical))
self.helpers.set_value(
"NumSymbols", self.helpers["NumSymbols"](self.X, self.y,
self.categorical))
self.helpers.set_value("ClassOccurences",
self.helpers["ClassOccurences"](self.X, self.y))
self.helpers.set_value(
"Skewnesses",
self.helpers["Skewnesses"](self.X_transformed, self.y,
self.categorical_transformed))
self.helpers.set_value(
"Kurtosisses",
self.helpers["Kurtosisses"](self.X_transformed, self.y,
self.categorical_transformed))
def calculate_all_metafeatures(X,
y,
categorical,
dataset_name,
calculate=None,
dont_calculate=None,
densify_threshold=1000):
logger = get_logger(__name__)
"""Calculate all metafeatures."""
helper_functions.clear()
metafeatures.clear()
mf_ = dict()
visited = set()
to_visit = deque()
to_visit.extend(metafeatures)
X_transformed = None
y_transformed = None
# TODO calculate the numpy metafeatures after all others to consume less
# memory
while len(to_visit) > 0:
name = to_visit.pop()
if calculate is not None and name not in calculate:
continue
if dont_calculate is not None and name in dont_calculate:
continue
if name in npy_metafeatures:
if X_transformed is None:
# TODO make sure this is done as efficient as possible (no copy for
# sparse matrices because of wrong sparse format)
sparse = scipy.sparse.issparse(X)
ohe = OneHotEncoder(categorical_features=categorical,
sparse=True)
X_transformed = ohe.fit_transform(X)
imputer = Imputer(strategy='mean', copy=False, dtype=X.dtype)
X_transformed = imputer.fit_transform(X_transformed)
standard_scaler = StandardScaler(copy=False)
X_transformed = standard_scaler.fit_transform(X_transformed)
# Transform the array which indicates the categorical metafeatures
number_numerical = np.sum(~np.array(categorical))
categorical_transformed = [True] * (X_transformed.shape[1] -
number_numerical) + \
[False] * number_numerical
# Densify the transformed matrix
if not sparse and scipy.sparse.issparse(X_transformed):
bytes_per_float = X_transformed.dtype.itemsize
num_elements = X_transformed.shape[
0] * X_transformed.shape[1]
megabytes_required = num_elements * bytes_per_float / 1000 / 1000
if megabytes_required < densify_threshold:
X_transformed = X_transformed.todense()
# This is not only important for datasets which are somehow
# sorted in a strange way, but also prevents lda from failing in
# some cases.
# Because this is advanced indexing, a copy of the data is returned!!!
X_transformed = check_array(X_transformed,
force_all_finite=True,
accept_sparse='csr')
rs = np.random.RandomState(42)
indices = np.arange(X_transformed.shape[0])
rs.shuffle(indices)
# TODO Shuffle inplace
X_transformed = X_transformed[indices]
y_transformed = y[indices]
X_ = X_transformed
y_ = y_transformed
categorical_ = categorical_transformed
else:
X_ = X
y_ = y
categorical_ = categorical
dependency = metafeatures.get_dependency(name)
if dependency is not None:
is_metafeature = dependency in metafeatures
is_helper_function = dependency in helper_functions
if is_metafeature and is_helper_function:
raise NotImplementedError()
elif not is_metafeature and not is_helper_function:
raise ValueError(dependency)
elif is_metafeature and not metafeatures.is_calculated(dependency):
to_visit.appendleft(name)
continue
elif is_helper_function and not helper_functions.is_calculated(
dependency):
logger.info("%s: Going to calculate: %s", dataset_name,
dependency)
value = helper_functions[dependency](X_, y_, categorical_)
helper_functions.set_value(dependency, value)
mf_[dependency] = value
logger.info("%s: Going to calculate: %s", dataset_name, name)
value = metafeatures[name](X_, y_, categorical_)
metafeatures.set_value(name, value)
mf_[name] = value
visited.add(name)
mf_ = DatasetMetafeatures(dataset_name, mf_)
return mf_
def calculate_all_metafeatures(X, y, categorical, dataset_name,
calculate=None, dont_calculate=None, densify_threshold=1000):
logger = get_logger(__name__)
"""Calculate all metafeatures."""
helper_functions.clear()
metafeatures.clear()
mf_ = dict()
visited = set()
to_visit = deque()
to_visit.extend(metafeatures)
X_transformed = None
y_transformed = None
# TODO calculate the numpy metafeatures after all others to consume less
# memory
while len(to_visit) > 0:
name = to_visit.pop()
if calculate is not None and name not in calculate:
continue
if dont_calculate is not None and name in dont_calculate:
continue
if name in npy_metafeatures:
if X_transformed is None:
# TODO make sure this is done as efficient as possible (no copy for
# sparse matrices because of wrong sparse format)
sparse = scipy.sparse.issparse(X)
ohe = OneHotEncoder(categorical_features=categorical, sparse=True)
X_transformed = ohe.fit_transform(X)
imputer = Imputer(strategy='mean', copy=False, dtype=X.dtype)
X_transformed = imputer.fit_transform(X_transformed)
standard_scaler = StandardScaler(copy=False)
X_transformed = standard_scaler.fit_transform(X_transformed)
# Transform the array which indicates the categorical metafeatures
number_numerical = np.sum(~np.array(categorical))
categorical_transformed = [True] * (X_transformed.shape[1] -
number_numerical) + \
[False] * number_numerical
# Densify the transformed matrix
if not sparse and scipy.sparse.issparse(X_transformed):
bytes_per_float = X_transformed.dtype.itemsize
num_elements = X_transformed.shape[0] * X_transformed.shape[1]
megabytes_required = num_elements * bytes_per_float / 1000 / 1000
if megabytes_required < densify_threshold:
X_transformed = X_transformed.todense()
# This is not only important for datasets which are somehow
# sorted in a strange way, but also prevents lda from failing in
# some cases.
# Because this is advanced indexing, a copy of the data is returned!!!
X_transformed = check_array(X_transformed,
force_all_finite=True,
accept_sparse='csr')
rs = np.random.RandomState(42)
indices = np.arange(X_transformed.shape[0])
rs.shuffle(indices)
# TODO Shuffle inplace
X_transformed = X_transformed[indices]
y_transformed = y[indices]
X_ = X_transformed
y_ = y_transformed
categorical_ = categorical_transformed
else:
X_ = X
y_ = y
categorical_ = categorical
dependency = metafeatures.get_dependency(name)
if dependency is not None:
is_metafeature = dependency in metafeatures
is_helper_function = dependency in helper_functions
if is_metafeature and is_helper_function:
raise NotImplementedError()
elif not is_metafeature and not is_helper_function:
raise ValueError(dependency)
elif is_metafeature and not metafeatures.is_calculated(dependency):
to_visit.appendleft(name)
continue
elif is_helper_function and not helper_functions.is_calculated(
dependency):
logger.info("%s: Going to calculate: %s", dataset_name,
dependency)
value = helper_functions[dependency](X_, y_, categorical_)
helper_functions.set_value(dependency, value)
mf_[dependency] = value
logger.info("%s: Going to calculate: %s", dataset_name,
name)
value = metafeatures[name](X_, y_, categorical_)
metafeatures.set_value(name, value)
mf_[name] = value
visited.add(name)
mf_ = DatasetMetafeatures(dataset_name, mf_)
return mf_
def __init__(self, random_state):
from autosklearn.pipeline.implementations.StandardScaler import \
StandardScaler
self.preprocessor = StandardScaler()
def test_scaler_2d_arrays(self):
"""Test scaling of 2d array along first axis"""
rng = np.random.RandomState(0)
X = rng.randn(4, 5)
X[:, 0] = 0.0 # first feature is always of zero
scaler = StandardScaler()
X_scaled = scaler.fit(X).transform(X, copy=True)
self.assertFalse(np.any(np.isnan(X_scaled)))
assert_array_almost_equal(X_scaled.mean(axis=0), 5 * [0.0])
assert_array_almost_equal(X_scaled.std(axis=0), [0., 1., 1., 1., 1.])
# Check that X has been copied
self.assertTrue(X_scaled is not X)
# check inverse transform
X_scaled_back = scaler.inverse_transform(X_scaled)
self.assertTrue(X_scaled_back is not X)
self.assertTrue(X_scaled_back is not X_scaled)
assert_array_almost_equal(X_scaled_back, X)
X_scaled = scale(X, axis=1, with_std=False)
self.assertFalse(np.any(np.isnan(X_scaled)))
assert_array_almost_equal(X_scaled.mean(axis=1), 4 * [0.0])
X_scaled = scale(X, axis=1, with_std=True)
self.assertFalse(np.any(np.isnan(X_scaled)))
assert_array_almost_equal(X_scaled.mean(axis=1), 4 * [0.0])
assert_array_almost_equal(X_scaled.std(axis=1), 4 * [1.0])
# Check that the data hasn't been modified
self.assertTrue(X_scaled is not X)
X_scaled = scaler.fit(X).transform(X, copy=False)
self.assertFalse(np.any(np.isnan(X_scaled)))
assert_array_almost_equal(X_scaled.mean(axis=0), 5 * [0.0])
assert_array_almost_equal(X_scaled.std(axis=0), [0., 1., 1., 1., 1.])
# Check that X has not been copied
self.assertTrue(X_scaled is X)
X = rng.randn(4, 5)
X[:, 0] = 1.0 # first feature is a constant, non zero feature
scaler = StandardScaler()
X_scaled = scaler.fit(X).transform(X, copy=True)
self.assertFalse(np.any(np.isnan(X_scaled)))
assert_array_almost_equal(X_scaled.mean(axis=0), 5 * [0.0])
assert_array_almost_equal(X_scaled.std(axis=0), [0., 1., 1., 1., 1.])
# Check that X has not been copied
self.assertTrue(X_scaled is not X)
# Same thing for sparse matrices...
X = scipy.sparse.coo_matrix((np.random.random((12,)),
([i for i in range(12)],
[int(i / 3) for i in range(12)])))
X = X.tocsr()
scaler = StandardScaler()
X_scaled = scaler.fit(X).transform(X, copy=False)
self.assertFalse(np.any(np.isnan(X_scaled.data)))
assert_array_almost_equal(
[X_scaled.data[X_scaled.indptr[i]:X_scaled.indptr[i + 1]].mean()
for i in range(X_scaled.shape[1])],
np.zeros((4, ), dtype=np.float64))
assert_array_almost_equal(np.sqrt([
X_scaled.data[X_scaled.indptr[i]:X_scaled.indptr[i + 1]].var()
for i in range(X_scaled.shape[1])]),
np.ones((4, ), dtype=np.float64))
# Because we change the sparse format to csc, we cannot assert that
# the matrix did not change!
# self.assertTrue(X_scaled is X)
# Check that the matrix is still sparse
self.assertEqual(len(X.indices), 12)