def test_wrapped_not_picklable_learner_with_reduce(self):
sim1 = LambdaSimulation(5, lambda i: i, lambda i, c: [0, 1, 2],
lambda i, c, a: cast(float, a))
learner = WrappedLearner(NotPicklableLearnerWithReduce())
experiment = Experiment([sim1], [learner])
experiment.evaluate()
def test_wrapped_not_picklable_learner_sans_reduce(self):
sim1 = LambdaSimulation(5, lambda i: i, lambda i, c: [0, 1, 2],
lambda i, c, a: cast(float, a))
learner = WrappedLearner(NotPicklableLearner())
experiment = Experiment([sim1], [learner])
CobaContext.logger = BasicLogger(ListSink())
experiment.evaluate()
self.assertEqual(1, len(CobaContext.logger.sink.items))
self.assertIn("pickle", CobaContext.logger.sink.items[0])
def test_no_params(self):
sim1 = NoParamsEnvironment()
learner = NoParamsLearner()
experiment = Experiment([sim1], [learner],
evaluation_task=OnlineOnPolicyEvalTask(False))
result = experiment.evaluate()
actual_learners = result.learners.to_tuples()
actual_environments = result.environments.to_tuples()
actual_interactions = result.interactions.to_tuples()
expected_learners = [(0, 'NoParamsLearner', 'NoParamsLearner')]
expected_environments = [(0, 'NoParamsEnvironment')]
expected_interactions = [(0, 0, 1, 0), (0, 0, 2, 1)]
self.assertCountEqual(actual_learners, expected_learners)
self.assertCountEqual(actual_environments, expected_environments)
self.assertCountEqual(actual_interactions, expected_interactions)
def test_sim(self):
sim1 = LambdaSimulation(2, lambda i: i, lambda i, c: [0, 1, 2],
lambda i, c, a: cast(float, a))
learner = ModuloLearner()
experiment = Experiment([sim1], [learner],
evaluation_task=OnlineOnPolicyEvalTask(False))
result = experiment.evaluate()
actual_learners = result.learners.to_tuples()
actual_environments = result.environments.to_tuples()
actual_interactions = result.interactions.to_tuples()
expected_learners = [(0, "Modulo", "Modulo(p=0)", '0')]
expected_environments = [(0, 'LambdaSimulation')]
expected_interactions = [(0, 0, 1, 0), (0, 0, 2, 1)]
self.assertCountEqual(actual_learners, expected_learners)
self.assertCountEqual(actual_environments, expected_environments)
self.assertCountEqual(actual_interactions, expected_interactions)
def test_learners(self):
sim = LambdaSimulation(2, lambda i: i, lambda i, c: [0, 1, 2],
lambda i, c, a: cast(float, a))
learner1 = ModuloLearner("0") #type: ignore
learner2 = ModuloLearner("1") #type: ignore
experiment = Experiment([sim], [learner1, learner2],
evaluation_task=OnlineOnPolicyEvalTask(False))
actual_result = experiment.evaluate()
actual_learners = actual_result._learners.to_tuples()
actual_environments = actual_result._environments.to_tuples()
actual_interactions = actual_result.interactions.to_tuples()
expected_learners = [(0, "Modulo", "Modulo(p=0)", '0'),
(1, "Modulo", "Modulo(p=1)", '1')]
expected_environments = [(0, 'LambdaSimulation')]
expected_interactions = [(0, 0, 1, 0), (0, 0, 2, 1), (0, 1, 1, 0),
(0, 1, 2, 1)]
self.assertCountEqual(actual_learners, expected_learners)
self.assertCountEqual(actual_environments, expected_environments)
self.assertCountEqual(actual_interactions, expected_interactions)
def test_ignore_raise(self):
CobaContext.logger = IndentLogger(ListSink())
sim1 = LambdaSimulation(2, lambda i: i, lambda i, c: [0, 1, 2],
lambda i, c, a: cast(float, a))
sim2 = LambdaSimulation(3, lambda i: i, lambda i, c: [3, 4, 5],
lambda i, c, a: cast(float, a))
experiment = Experiment(
[sim1, sim2], [ModuloLearner(), BrokenLearner()],
evaluation_task=OnlineOnPolicyEvalTask(False))
result = experiment.evaluate()
actual_learners = result.learners.to_tuples()
actual_environments = result.environments.to_tuples()
actual_interactions = result.interactions.to_tuples()
expected_learners = [(0, "Modulo", "Modulo(p=0)", '0'),
(1, "Broken", "Broken", float('nan'))]
expected_environments = [(0, 'LambdaSimulation'),
(1, 'LambdaSimulation')]
expected_interactions = [(0, 0, 1, 0), (0, 0, 2, 1), (1, 0, 1, 3),
(1, 0, 2, 4), (1, 0, 3, 5)]
self.assertIsInstance(CobaContext.logger, IndentLogger)
self.assertEqual(
2,
sum([
int("Unexpected exception:" in item)
for item in CobaContext.logger.sink.items
]))
self.assertCountEqual(actual_learners[0], expected_learners[0])
self.assertCountEqual(actual_learners[1][:3], expected_learners[1][:3])
self.assertTrue(math.isnan(expected_learners[1][3]))
self.assertCountEqual(actual_environments, expected_environments)
self.assertCountEqual(actual_interactions, expected_interactions)