def __call__(self, data):
method = self.method
# select default method according to the provided data
if method is None:
autoMethod = True
discr_ratio = (sum(a.is_discrete for a in data.domain.attributes) /
len(data.domain.attributes))
if data.domain.has_discrete_class:
if discr_ratio >= 0.5:
method = GainRatio()
else:
method = ANOVA()
else:
method = UnivariateLinearRegression()
features = data.domain.attributes
try:
scores = method(data)
except ValueError:
scores = self.score_only_nice_features(data, method)
best = sorted(zip(scores, features),
key=itemgetter(0),
reverse=self.decreasing)
if self.k:
best = best[:self.k]
if self.threshold:
pred = ((lambda x: x[0] >= self.threshold) if self.decreasing else
(lambda x: x[0] <= self.threshold))
best = takewhile(pred, best)
domain = Orange.data.Domain([f for s, f in best],
data.domain.class_vars, data.domain.metas)
return data.transform(domain)
def test_select_threshold(self):
data = Table('wine')
anova = ANOVA()
t = 30
data2 = SelectBestFeatures(method=anova, threshold=t)(data)
self.assertTrue(
all(anova(data, f) >= t for f in data2.domain.attributes))
def test_select_threshold(self):
anova = ANOVA()
t = 30
data2 = SelectBestFeatures(method=anova,
threshold=t)(self.heart_disease)
self.assertTrue(
all(
anova(self.heart_disease, f) >= t
for f in data2.domain.attributes))
def test_wrong_class_type(self):
scorers = [Gini(), InfoGain(), GainRatio()]
for scorer in scorers:
with self.assertRaises(ValueError):
scorer(self.housing, 0)
with self.assertRaises(ValueError):
Chi2()(self.housing, 0)
with self.assertRaises(ValueError):
ANOVA()(self.housing, 2)
UnivariateLinearRegression()(self.housing, 2)
def test_anova(self):
nrows, ncols = 500, 5
X = np.random.rand(nrows, ncols)
y = 4 + (-3 * X[:, 1] + X[:, 3]) // 2
domain = Domain.from_numpy(X, y)
domain = Domain(domain.attributes,
DiscreteVariable('c', values=np.unique(y)))
data = Table(domain, X, y)
scorer = ANOVA()
sc = [scorer(data, a) for a in range(ncols)]
self.assertTrue(np.argmax(sc) == 1)
def __call__(self, data):
method = self.method
# select default method according to the provided data
if method is None:
autoMethod = True
discr_ratio = (sum(a.is_discrete for a in data.domain.attributes) /
len(data.domain.attributes))
if data.domain.has_discrete_class:
if discr_ratio >= 0.5:
method = GainRatio()
else:
method = ANOVA()
else:
method = UnivariateLinearRegression()
if not isinstance(data.domain.class_var, method.class_type):
raise ValueError(
("Scoring method {} requires a class variable " +
"of type {}.").format((method if type(method) == type else
type(method)).__name__,
method.class_type.__name__))
features = data.domain.attributes
try:
scores = method(data)
except ValueError:
scores = self.score_only_nice_features(data, method)
best = sorted(zip(scores, features),
key=itemgetter(0),
reverse=self.decreasing)
if self.k:
best = best[:self.k]
if self.threshold:
pred = ((lambda x: x[0] >= self.threshold) if self.decreasing else
(lambda x: x[0] <= self.threshold))
best = takewhile(pred, best)
domain = Orange.data.Domain([f for s, f in best],
data.domain.class_vars, data.domain.metas)
return data.from_table(domain, data)
