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)
center = not scipy.sparse.isspmatrix((X_transformed))
standard_scaler = StandardScaler(with_mean=center)
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 perform1HotEncoding(self):
if not hasattr(self, 'data'):
raise ValueError('perform1HotEncoding can only be called when '
'data is loaded')
if hasattr(self, 'encoder_'):
raise ValueError('perform1HotEncoding can only be called on '
'non-encoded data.')
self._encoder = None
sparse = True if self.info['is_sparse'] == 1 else False
has_missing = True if self.info['has_missing'] else False
to_encode = ['categorical']
if has_missing:
to_encode += ['binary']
encoding_mask = [
feat_type.lower() in to_encode for feat_type in self.feat_type
]
categorical = [
True if feat_type.lower() == 'categorical' else False
for feat_type in self.feat_type
]
predicted_RAM_usage = float(
predict_RAM_usage(self.data['X_train'], categorical)) / 1024 / 1024
if predicted_RAM_usage > 1000:
sparse = True
if any(encoding_mask):
encoder = OneHotEncoder(categorical_features=encoding_mask,
dtype=np.float32,
sparse=sparse)
self.data['X_train'] = encoder.fit_transform(self.data['X_train'])
if 'X_valid' in self.data:
self.data['X_valid'] = encoder.transform(self.data['X_valid'])
if 'X_test' in self.data:
self.data['X_test'] = encoder.transform(self.data['X_test'])
if not sparse and scipy.sparse.issparse(self.data['X_train']):
self.data['X_train'] = self.data['X_train'].todense()
if 'X_valid' in self.data:
self.data['X_valid'] = self.data['X_valid'].todense()
if 'X_test' in self.data:
self.data['X_test'] = self.data['X_test'].todense()
self.encoder = encoder
self.info['is_sparse'] = 1 if sparse else 0
def test_transform_with_unknown_value(self):
input = np.array(((0, 1, 2, 3, 4, 5), (0, 1, 2, 3, 4, 5))).transpose()
ohe = OneHotEncoder()
ohe.fit(input)
test_data = np.array(((0, 1, 2, 6), (0, 1, 6, 7))).transpose()
output = ohe.transform(test_data).todense()
self.assertEqual(5, np.sum(output))
input = np.array(((0, 1, 2, 3, 4, 5), (0, 1, 2, 3, 4, 5))).transpose()
ips = scipy.sparse.csr_matrix(input)
ohe = OneHotEncoder()
ohe.fit(ips)
test_data = np.array(((0, 1, 2, 6), (0, 1, 6, 7))).transpose()
tds = scipy.sparse.csr_matrix(test_data)
output = ohe.transform(tds).todense()
self.assertEqual(3, np.sum(output))
def perform1HotEncoding(self):
if not hasattr(self, 'data'):
raise ValueError('perform1HotEncoding can only be called when '
'data is loaded')
if hasattr(self, 'encoder_'):
raise ValueError('perform1HotEncoding can only be called on '
'non-encoded data.')
self._encoder = None
sparse = True if self.info['is_sparse'] == 1 else False
has_missing = True if self.info['has_missing'] else False
to_encode = ['categorical']
if has_missing:
to_encode += ['binary']
encoding_mask = [feat_type.lower() in to_encode
for feat_type in self.feat_type]
categorical = [True if feat_type.lower() == 'categorical' else False
for feat_type in self.feat_type]
predicted_RAM_usage = float(predict_RAM_usage(
self.data['X_train'], categorical)) / 1024 / 1024
if predicted_RAM_usage > 1000:
sparse = True
if any(encoding_mask):
encoder = OneHotEncoder(categorical_features=encoding_mask,
dtype=np.float32,
sparse=sparse)
self.data['X_train'] = encoder.fit_transform(self.data['X_train'])
if 'X_valid' in self.data:
self.data['X_valid'] = encoder.transform(self.data['X_valid'])
if 'X_test' in self.data:
self.data['X_test'] = encoder.transform(self.data['X_test'])
if not sparse and scipy.sparse.issparse(self.data['X_train']):
self.data['X_train'] = self.data['X_train'].todense()
if 'X_valid' in self.data:
self.data['X_valid'] = self.data['X_valid'].todense()
if 'X_test' in self.data:
self.data['X_test'] = self.data['X_test'].todense()
self.encoder = encoder
self.info['is_sparse'] = 1 if sparse else 0
def fit_then_transform_dense(self, expected, input,
categorical_features='all',
minimum_fraction=None):
ohe = OneHotEncoder(categorical_features=categorical_features,
sparse=False, minimum_fraction=minimum_fraction)
transformation = ohe.fit_transform(input.copy())
self.assertIsInstance(transformation, np.ndarray)
assert_array_almost_equal(expected, transformation)
ohe2 = OneHotEncoder(categorical_features=categorical_features,
sparse=False, minimum_fraction=minimum_fraction)
ohe2.fit(input.copy())
transformation = ohe2.transform(input.copy())
self.assertIsInstance(transformation, np.ndarray)
assert_array_almost_equal(expected, transformation)
def perform_one_hot_encoding(sparse, categorical, data):
predicted_RAM_usage = float(predict_RAM_usage(data[0], categorical)) / 1024 / 1024
if predicted_RAM_usage > 1000:
sparse = True
rvals = []
if any(categorical):
encoder = OneHotEncoder(categorical_features=categorical,
dtype=np.float32,
sparse=sparse)
rvals.append(encoder.fit_transform(data[0]))
for d in data[1:]:
rvals.append(encoder.transform(d))
if not sparse and scipy.sparse.issparse(rvals[0]):
for i in range(len(rvals)):
rvals[i] = rvals[i].todense()
else:
rvals = data
return rvals, sparse
def perform_one_hot_encoding(sparse, categorical, data):
predicted_RAM_usage = float(
predict_RAM_usage(data[0], categorical)) / 1024 / 1024
if predicted_RAM_usage > 1000:
sparse = True
rvals = []
if any(categorical):
encoder = OneHotEncoder(categorical_features=categorical,
dtype=np.float32,
sparse=sparse)
rvals.append(encoder.fit_transform(data[0]))
for d in data[1:]:
rvals.append(encoder.transform(d))
if not sparse and scipy.sparse.issparse(rvals[0]):
for i in range(len(rvals)):
rvals[i] = rvals[i].todense()
else:
rvals = data
return rvals, sparse
def fit_then_transform(self, expected, input, categorical_features='all',
minimum_fraction=None):
# Test fit_transform
ohe = OneHotEncoder(categorical_features=categorical_features,
minimum_fraction=minimum_fraction)
transformation = ohe.fit_transform(input.copy())
self.assertIsInstance(transformation, scipy.sparse.csr_matrix)
assert_array_almost_equal(expected.astype(float),
transformation.todense())
# Test fit, and afterwards transform
ohe2 = OneHotEncoder(categorical_features=categorical_features,
minimum_fraction=minimum_fraction)
ohe2.fit(input.copy())
transformation = ohe2.transform(input.copy())
self.assertIsInstance(transformation, scipy.sparse.csr_matrix)
assert_array_almost_equal(expected, transformation.todense())
def test_classification_workflow(self):
task = openml.tasks.get_task(254)
X, y = task.get_X_and_y()
ohe = OneHotEncoder(categorical_features=[True] * 22)
tree = sklearn.tree.DecisionTreeClassifier(random_state=1)
pipeline = sklearn.pipeline.Pipeline((('ohe', ohe), ('tree', tree)))
X_train, X_test, y_train, y_test = \
sklearn.model_selection.train_test_split(X, y, random_state=3,
train_size=0.5,
test_size=0.5)
pipeline.fit(X_train, y_train)
self.assertEqual(np.mean(y_train == pipeline.predict(X_train)), 1)
# With an incorrect copy operation the OneHotEncoder would rearrange
# the data in such a way that the accuracy would drop to 66%
self.assertEqual(np.mean(y_test == pipeline.predict(X_test)), 1)
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, ))
# First, swap NaNs and zeros, because when converting an encoded
# dense matrix to sparse, the values which are encoded to zero are lost
X_sparse = X.copy()
NaNs = ~np.isfinite(X_sparse)
X_sparse[NaNs] = 0
X_sparse = sparse.csr_matrix(X_sparse)
ohe = OneHotEncoder(self.categorical)
X_transformed = X_sparse.copy()
X_transformed = ohe.fit_transform(X_transformed)
imp = SimpleImputer(copy=False)
X_transformed = imp.fit_transform(X_transformed)
standard_scaler = StandardScaler(with_mean=False)
X_transformed = standard_scaler.fit_transform(X_transformed)
# 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_sparse
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.mf.set_value(
"NumberOfMissingValues",
self.mf["NumberOfMissingValues"](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 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,))
# First, swap NaNs and zeros, because when converting an encoded
# dense matrix to sparse, the values which are encoded to zero are lost
X_sparse = X.copy()
NaNs = ~np.isfinite(X_sparse)
X_sparse[NaNs] = 0
X_sparse = sparse.csr_matrix(X_sparse)
ohe = OneHotEncoder(self.categorical)
X_transformed = X_sparse.copy()
X_transformed = ohe.fit_transform(X_transformed)
imp = Imputer(copy=False)
X_transformed = imp.fit_transform(X_transformed)
standard_scaler = StandardScaler()
X_transformed = standard_scaler.fit_transform(X_transformed)
# 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_sparse
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.mf.set_value("NumberOfMissingValues",
self.mf["NumberOfMissingValues"](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_