def test_simple_openml_source(self) -> None:
#this test requires interet acess to download the data
ExecutionContext.FileCache = NoneCache()
simulation = ClassificationSimulation(*OpenmlClassificationSource(1116).read())
#simulation = ClassificationSimulation.from_source(OpenmlSource(273))
self.assertEqual(len(simulation.interactions), 6598)
for rnd in simulation.interactions:
hash(rnd.context) #make sure these are hashable
hash(rnd.actions[0]) #make sure these are hashable
hash(rnd.actions[1]) #make sure these are hashable
self.assertEqual(len(cast(Tuple,rnd.context)), 268)
self.assertIn((1,0), rnd.actions)
self.assertIn((0,1), rnd.actions)
self.assertEqual(len(rnd.actions),2)
actual_rewards = simulation.reward(_choices(rnd))
self.assertIn(1, actual_rewards)
self.assertIn(0, actual_rewards)
def test_large_from_openml(self) -> None:
#this test requires interet acess to download the data
ExecutionContext.FileCache = MemoryCache()
OpenmlClassificationSource(154).read() #this will cause it to read and cache in memory so we don't measure read time
time = min(timeit.repeat(lambda:ClassificationSimulation(*OpenmlClassificationSource(154).read()), repeat=1, number=1))
print(time)
#with caching took approximately 17 seconds to encode
self.assertLess(time, 30)
def test_sparse(self) -> None:
feature_rows = [((0, 1), (10, 11)), ((1, 2), (20, 30)),
((2, 3), (30, 40)), ((2, 3), (30, 40))]
label_column = (1, 1, 0, 2)
simulation = ClassificationSimulation(feature_rows, label_column)
interactions = list(simulation.read())
self.assertEqual(dict(zip(*feature_rows[0])), interactions[0].context)
self.assertEqual(dict(zip(*feature_rows[1])), interactions[1].context)
self.assertEqual(dict(zip(*feature_rows[2])), interactions[2].context)
self.assertEqual(dict(zip(*feature_rows[3])), interactions[3].context)
self.assertEqual([1, 0, 2], interactions[0].actions)
self.assertEqual([1, 0, 2], interactions[1].actions)
self.assertEqual([1, 0, 2], interactions[2].actions)
self.assertEqual([1, 0, 2], interactions[3].actions)
self.assertEqual([1, 0, 0], interactions[0].feedbacks)
self.assertEqual([1, 0, 0], interactions[1].feedbacks)
self.assertEqual([0, 1, 0], interactions[2].feedbacks)
self.assertEqual([0, 0, 1], interactions[3].feedbacks)
def test_constructor_with_too_few_labels(self) -> None:
with self.assertRaises(AssertionError):
ClassificationSimulation([1, 1], [1])
def test_constructor_with_good_features_and_labels3(self) -> None:
features = [(1, 2), (3, 4)]
labels = ["good", "bad"]
simulation = ClassificationSimulation(features, labels)
self.assert_simulation_for_data(simulation, features, labels)
def test_constructor_with_good_features_and_labels1(self) -> None:
features = [1, 2, 3, 4]
labels = [1, 1, 0, 0]
simulation = ClassificationSimulation(features, labels)
self.assert_simulation_for_data(simulation, features, labels)