def test_value_from_probs(self):
nrows = 100
ncols = 5
x = np.random.randint(0, 2, (nrows, ncols))
# single class variable
y = np.random.randint(1, 4, (nrows, 1)) // 2 # majority = 1
t = Table(x, y)
learn = DummyLearner()
clf = learn(t)
clf.ret = Model.Probs
y2 = clf(x, ret=Model.Value)
self.assertEqual(y2.shape, (nrows, ))
y2, probs = clf(x, ret=Model.ValueProbs)
self.assertEqual(y2.shape, (nrows, ))
self.assertEqual(probs.shape, (nrows, 2))
# multitarget
y = np.random.randint(1, 6, (nrows, 2))
y[:, 0] = y[:, 0] // 3 # majority = 1
y[:, 1] = (y[:, 1] + 4) // 3 # majority = 2
domain = Domain(
[ContinuousVariable('i' + str(i)) for i in range(ncols)], [
DiscreteVariable('c' + str(i), values="0123")
for i in range(y.shape[1])
])
t = Table(domain, x, y)
learn = DummyMulticlassLearner()
clf = learn(t)
clf.ret = Model.Probs
y2 = clf(x, ret=Model.Value)
self.assertEqual(y2.shape, y.shape)
y2, probs = clf(x, ret=Model.ValueProbs)
self.assertEqual(y2.shape, y.shape)
self.assertEqual(probs.shape, (nrows, 2, 4))
def test_probs_from_value(self):
nrows = 100
ncols = 5
x = np.random.random_integers(0, 1, (nrows, ncols))
# single class variable
y = np.random.random_integers(1, 2, (nrows, 1))
t = data.Table(x, y)
learn = DummyLearner()
clf = learn(t)
clf.ret = Orange.classification.Model.Value
y2 = clf(x, ret=Orange.classification.Model.Probs)
self.assertTrue(y2.shape == (nrows, 3))
y2, probs = clf(x, ret=Orange.classification.Model.ValueProbs)
self.assertTrue(y2.shape == (nrows, ))
self.assertTrue(probs.shape == (nrows, 3))
# multitarget
y = np.random.random_integers(1, 5, (nrows, 2))
y[:, 0] = y[:, 0] // 3 # majority = 1
y[:, 1] = (y[:, 1] + 4) // 3 # majority = 2
t = data.Table(x, y)
learn = DummyMulticlassLearner()
clf = learn(t)
clf.ret = Orange.classification.Model.Value
probs = clf(x, ret=Orange.classification.Model.Probs)
self.assertEqual(probs.shape, (nrows, 2, 4))
y2, probs = clf(x, ret=Orange.classification.Model.ValueProbs)
self.assertEqual(y2.shape, y.shape)
self.assertEqual(probs.shape, (nrows, 2, 4))
def test_unsupported(self):
nrows = 20
ncols = 10
x = np.random.random_integers(1, 3, (nrows, ncols))
# multiple class variables
y = np.random.random_integers(10, 11, (nrows, 2))
t = Table(x, y)
learn = DummyLearner()
with self.assertRaises(TypeError):
clf = learn(t)
# single class variable
y = np.random.random_integers(10, 11, (nrows, 1))
t = Table(x, y)
learn = DummyLearner()
clf = learn(t)
z = clf(x)
self.assertEqual(z.ndim, 1)
def test_unsupported(self):
nrows = 20
ncols = 10
x = np.random.randint(1, 4, (nrows, ncols))
# multiple class variables
y = np.random.randint(0, 2, (nrows, 2))
t = Table(x, y)
learn = DummyLearner()
# TODO: Errors raised from various data checks should be made consistent
with self.assertRaises((ValueError, TypeError)):
clf = learn(t)
# single class variable
y = np.random.randint(0, 2, (nrows, 1))
t = Table(x, y)
learn = DummyLearner()
clf = learn(t)
z = clf(x)
self.assertEqual(z.ndim, 1)
def test_single_class(self):
rows = 10
attr = 3
vars = 1
class_var_domain = 20
self.prepareTable(rows, attr, vars, class_var_domain)
y = np.random.randint(2, 6, (rows, vars)) * 2
t = Table(self.domain, self.x, y)
learn = DummyLearner()
clf = learn(t)
z, p = clf(self.x, ret=Model.ValueProbs)
self.assertEqual(p.shape, (rows, class_var_domain))
self.assertTrue(np.all(z == np.argmax(p, axis=-1)))
def test_probs_from_value(self):
nrows = 100
ncols = 5
x = np.random.randint(0, 2, (nrows, ncols))
# single class variable
y = np.random.randint(0, 2, (nrows, 1))
d = Domain(
[
DiscreteVariable("v" + str(i),
values=[str(v) for v in np.unique(x[:, i])])
for i in range(ncols)
],
DiscreteVariable("c", values="12"),
)
t = Table(d, x, y)
learn = DummyLearner()
clf = learn(t)
clf.ret = Model.Value
y2 = clf(x, ret=Model.Probs)
self.assertEqual(y2.shape, (nrows, 2))
y2, probs = clf(x, ret=Model.ValueProbs)
self.assertEqual(y2.shape, (nrows, ))
self.assertEqual(probs.shape, (nrows, 2))
# multitarget
y = np.random.randint(1, 6, (nrows, 2))
y[:, 0] = y[:, 0] // 3 # majority = 1
y[:, 1] = (y[:, 1] + 4) // 3 - 1 # majority = 1
domain = Domain(
[ContinuousVariable("i" + str(i)) for i in range(ncols)],
[
DiscreteVariable("c" + str(i), values="0123")
for i in range(y.shape[1])
],
)
t = Table(domain, x, y)
learn = DummyMulticlassLearner()
clf = learn(t)
clf.ret = Model.Value
probs = clf(x, ret=Model.Probs)
self.assertEqual(probs.shape, (nrows, 2, 4))
y2, probs = clf(x, ret=Model.ValueProbs)
self.assertEqual(y2.shape, y.shape)
self.assertEqual(probs.shape, (nrows, 2, 4))
def test_probs_from_value(self):
nrows = 100
ncols = 5
x = np.random.random_integers(0, 1, (nrows, ncols))
# single class variable
y = np.random.random_integers(0, 1, (nrows, 1))
t = Table(
Domain([
DiscreteVariable('v' + str(i), values=np.unique(x[:, i]))
for i in range(ncols)
], DiscreteVariable('c', values=[1, 2])), x, y)
learn = DummyLearner()
clf = learn(t)
clf.ret = Model.Value
y2 = clf(x, ret=Model.Probs)
self.assertEqual(y2.shape, (nrows, 2))
y2, probs = clf(x, ret=Model.ValueProbs)
self.assertEqual(y2.shape, (nrows, ))
self.assertEqual(probs.shape, (nrows, 2))
# multitarget
y = np.random.random_integers(1, 5, (nrows, 2))
y[:, 0] = y[:, 0] // 3 # majority = 1
y[:, 1] = (y[:, 1] + 4) // 3 - 1 # majority = 1
domain = Domain(
[ContinuousVariable('i' + str(i)) for i in range(ncols)], [
DiscreteVariable('c' + str(i), values=range(4))
for i in range(y.shape[1])
])
t = Table(domain, x, y)
learn = DummyMulticlassLearner()
clf = learn(t)
clf.ret = Model.Value
probs = clf(x, ret=Model.Probs)
self.assertEqual(probs.shape, (nrows, 2, 4))
y2, probs = clf(x, ret=Model.ValueProbs)
self.assertEqual(y2.shape, y.shape)
self.assertEqual(probs.shape, (nrows, 2, 4))
def test_incompatible_domain(self):
iris = Table("iris")
titanic = Table("titanic")
clf = DummyLearner()(iris)
with self.assertRaises(DomainTransformationError):
clf(titanic)
