def test_two_environment_tasks_first_environment_is_empty(self):
src1 = LinearSyntheticSimulation(n_interactions=0)
src2 = LinearSyntheticSimulation(n_interactions=5)
task1 = ObserveTask()
task2 = ObserveTask()
items = [ WorkItem(0, None, src1, None, task1), WorkItem(1, None, src2, None, task2) ]
transactions = list(ProcessWorkItems().filter(items))
self.assertEqual(len(task1.observed), 0)
self.assertEqual(len(task2.observed[1]), 5)
def test_linear_synthetic(self):
time = timeit.timeit(
lambda: list(LinearSyntheticSimulation(100).read()), number=1)
#.22 was my final time
self.assertLess(time, 2.2)
def test_simple_no_context_action_features(self):
simulation = LinearSyntheticSimulation(500,
n_actions=2,
n_context_features=0,
n_action_features=4,
reward_features=["a", "xa"])
interactions = list(simulation.read())
self.assertEqual(500, len(interactions))
self.assertEqual(2, len(interactions[0].actions))
self.assertEqual(None, interactions[0].context)
self.assertEqual(4, len(interactions[0].actions[0]))
rewards = [r for i in interactions for r in i.kwargs['rewards']]
self.assertLess(max(rewards), 1.2)
self.assertGreater(max(rewards), .75)
self.assertLess(min(rewards), .25)
self.assertGreater(min(rewards), -.2)
self.assertGreater(.05, abs(.5 - sum(rewards) / len(rewards)))
def test_empty_environment_tasks(self):
src1 = LinearSyntheticSimulation(0)
task1 = ObserveTask()
items = [ WorkItem(0, None, src1, None, task1) ]
transactions = list(ProcessWorkItems().filter(items))
self.assertEqual(len(task1.observed), 0)
self.assertEqual(len(transactions) , 0)
def test_pickle(self):
simulation = LinearSyntheticSimulation(500,
n_actions=2,
n_context_features=3,
n_action_features=4,
reward_features=["a", "xa"],
seed=2)
simulation = pickle.loads(pickle.dumps(simulation))
interactions = list(simulation.read())
self.assertEqual(500, len(interactions))
self.assertEqual(2, len(interactions[0].actions))
self.assertEqual(3, len(interactions[0].context))
self.assertEqual(4, len(interactions[0].actions[0]))
self.assertEqual(['a', 'xa'], simulation.params['reward_features'])
self.assertEqual(2, simulation.params['seed'])
self.assertEqual("LinearSynth(A=2,c=3,a=4,R=['a', 'xa'],seed=2)",
str(simulation))
def test_pipe_four_groups(self):
sim1 = Environments(LinearSyntheticSimulation(500)).shuffle([1,2])._environments
sim2 = LambdaSimulation(5, lambda i: i, lambda i,c: [0,1,2], lambda i,c,a: cast(float,a))
tasks = [
WorkItem(None, 0, None, None, None),
WorkItem(None, 1, None, None, None),
WorkItem(1, None, sim2, None, None),
WorkItem(0, None, sim1[0], None, None),
WorkItem(0, 0, sim1[0], None, None),
WorkItem(2, 0, sim1[1], None, None),
WorkItem(0, 1, sim1[0], None, None),
WorkItem(1, 1, sim2, None, None)
]
groups = list(ChunkBySource().filter(tasks))
self.assertEqual(len(groups), 4)
self.assertEqual(groups[0], tasks[0:1])
self.assertEqual(groups[1], tasks[1:2])
self.assertEqual(groups[2], [tasks[3],tasks[4],tasks[6],tasks[5]])
self.assertEqual(groups[3], [tasks[2],tasks[7]])
def test_str(self):
self.assertEqual(
"LinearSynth(A=2,c=3,a=4,R=['xa'],seed=2)",
str(LinearSyntheticSimulation(100, 2, 3, 4, ["xa"], 2)))
def test_params(self):
env = LinearSyntheticSimulation(100, reward_features=["xa"], seed=2)
self.assertEqual(['xa'], env.params['reward_features'])
self.assertEqual(2, env.params['seed'])