def test_relieff(self):
old_breast = self.breast.copy()
weights = ReliefF(random_state=42)(self.breast, None)
found = [
self.breast.domain[attr].name
for attr in reversed(weights.argsort()[-3:])
]
reference = ['Bare_Nuclei', 'Clump thickness', 'Marginal_Adhesion']
self.assertEqual(sorted(found), reference)
# Original data is unchanged
np.testing.assert_equal(old_breast.X, self.breast.X)
np.testing.assert_equal(old_breast.Y, self.breast.Y)
# Ensure it doesn't crash on adult dataset
weights = ReliefF(random_state=42)(self.lenses, None)
found = [
self.lenses.domain[attr].name for attr in weights.argsort()[-2:]
]
# some leeway for randomness in relieff random instance selection
self.assertIn('tear_rate', found)
# Ensure it doesn't crash on missing target class values
old_breast.Y[0] = np.nan
weights = ReliefF()(old_breast, None)
np.testing.assert_array_equal(
ReliefF(random_state=1)(self.breast, None),
ReliefF(random_state=1)(self.breast, None))
def test_relieff(self):
old_monk = self.monk.copy()
weights = ReliefF(random_state=42)(self.monk, None)
found = [
self.monk.domain[attr].name
for attr in reversed(weights.argsort()[-3:])
]
reference = ['a', 'b', 'e']
self.assertEqual(sorted(found), reference)
# Original data is unchanged
np.testing.assert_equal(old_monk.X, self.monk.X)
np.testing.assert_equal(old_monk.Y, self.monk.Y)
# Ensure it doesn't crash on adult dataset
weights = ReliefF(random_state=42)(self.adult, None)
found = [
self.adult.domain[attr].name for attr in weights.argsort()[-2:]
]
# some leeway for randomness in relieff random instance selection
self.assertIn('marital-status', found)
# Ensure it doesn't crash on missing target class values
old_monk.Y[0] = np.nan
weights = ReliefF()(old_monk, None)
np.testing.assert_array_equal(
ReliefF(random_state=1)(self.monk, None),
ReliefF(random_state=1)(self.monk, None))
def compute_attr_order(self):
"""
Order attributes by Relief if there is a target variable. In case of
ties or without target, order by name.
Add the class variable at the beginning when not coloring by class
distribution.
If `self.attrs` is not `None`, keep the ordering and just add or remove
the class as needed.
"""
data = self.master.discrete_data
class_var = data.domain.class_var
if not self.attr_ordering:
if class_var is None:
self.attr_ordering = sorted(data.domain, key=attrgetter("name"))
else:
weights = ReliefF(n_iterations=100, k_nearest=10)(data)
attrs = sorted(zip(weights, data.domain.attributes),
key=lambda x: (-x[0], x[1].name))
self.attr_ordering = [a for _, a in attrs]
if class_var is not None:
if self._compute_class_dists():
if self.attr_ordering[0] is class_var:
del self.attr_ordering[0]
elif self.attr_ordering[0] is not class_var:
self.attr_ordering.insert(0, class_var)
def test_relieff(self):
old_monk = self.monk.copy()
weights = ReliefF()(self.monk, None)
found = [self.monk.domain[attr].name for attr in reversed(weights.argsort()[-3:])]
reference = ['a', 'b', 'e']
self.assertEqual(sorted(found), reference)
# Original data is unchanged
np.testing.assert_equal(old_monk.X, self.monk.X)
np.testing.assert_equal(old_monk.Y, self.monk.Y)
# Ensure it doesn't crash on adult dataset
weights = ReliefF()(self.adult, None)
found = [self.adult.domain[attr].name for attr in weights.argsort()[-2:]]
# some leeway for randomness in relieff random instance selection
self.assertIn('marital-status', found)
# Ensure it doesn't crash on missing target class values
old_monk.Y[0] = np.nan
weights = ReliefF()(old_monk, None)
def _fit(self):
warnings.simplefilter(action='ignore')
table = self.data.to_table()
if self.params["eval_method"] == "relief":
scores = ReliefF(table, n_iterations=100)
elif self.params["eval_method"] == "fcbf":
scores = FCBF(table)
else:
scores = RandomForestLearner().score_data(table)[0]
for attr, score in zip(table.domain.attributes, scores):
self.feature_importances[attr.name] = score
def test_relieff(self):
old_breast = self.breast.copy()
weights = ReliefF(random_state=42)(self.breast, None)
found = [self.breast.domain[attr].name for attr in reversed(weights.argsort()[-3:])]
reference = ['Bare_Nuclei', 'Clump thickness', 'Marginal_Adhesion']
self.assertEqual(sorted(found), reference)
# Original data is unchanged
np.testing.assert_equal(old_breast.X, self.breast.X)
np.testing.assert_equal(old_breast.Y, self.breast.Y)
# Ensure it doesn't crash on adult dataset
weights = ReliefF(random_state=42)(self.lenses, None)
found = [self.lenses.domain[attr].name for attr in weights.argsort()[-2:]]
# some leeway for randomness in relieff random instance selection
self.assertIn('tear_rate', found)
# Ensure it doesn't crash on missing target class values
old_breast.Y[0] = np.nan
weights = ReliefF()(old_breast, None)
np.testing.assert_array_equal(
ReliefF(random_state=1)(self.breast, None),
ReliefF(random_state=1)(self.breast, None)
)
def score_heuristic(self):
X = self.parent_widget.graph.scaled_data.T
Y = self.parent_widget.data.Y
dom = Orange.data.Domain(
[ContinuousVariable(str(i)) for i in range(X.shape[1])],
self.parent_widget.data.domain.class_vars)
data = Orange.data.Table(dom, X, Y)
weights = ReliefF(n_iterations=100, k_nearest=self.k)(data)
attrs = sorted(zip(
weights,
(x.name for x in self.parent_widget.data.domain.attributes)),
reverse=True)
return [a for _, a in attrs]
def _compute_attr_order(self):
"""
used by VizRank to evaluate attributes
"""
master = self.master
attrs = [v for v in chain(master.model_selected[:], master.model_other[:])
if v is not self.attr_color]
data = self.master.data.transform(Domain(attributes=attrs, class_vars=self.attr_color))
self.data = data
self.valid_data = np.hstack((~np.isnan(data.X), ~np.isnan(data.Y.reshape(len(data.Y), 1))))
weights = ReliefF(n_iterations=100, k_nearest=self.minK)(data)
attrs = sorted(zip(weights, attrs), key=lambda x: (-x[0], x[1].name))
self.attr_ordering = attr_ordering = [a for _, a in attrs]
return attr_ordering
def test_relieff(self):
old_monk = self.monk.copy()
weights = ReliefF()(self.monk, None)
found = [
self.monk.domain[attr].name
for attr in reversed(weights.argsort()[-3:])
]
reference = ['a', 'b', 'e']
self.assertEqual(sorted(found), reference)
# Original data is unchanged
np.testing.assert_equal(old_monk.X, self.monk.X)
np.testing.assert_equal(old_monk.Y, self.monk.Y)
# Ensure it doesn't crash on adult dataset
weights = ReliefF()(self.adult, None)
found = sorted(
[self.adult.domain[attr].name for attr in weights.argsort()[-2:]])
reference = ['marital-status', 'relationship']
self.assertEqual(found, reference)
# Ensure it doesn't crash on missing target class values
old_monk.Y[0] = np.nan
weights = ReliefF()(old_monk, None)
