def test_from_linear_synthetic(self):
envs = Environments.from_linear_synthetic(100, 2, 3, 4, ["xa"], 5)
env = envs[0]
self.assertIsInstance(env, LinearSyntheticSimulation)
interactions = list(env.read())
self.assertEqual(1, len(envs))
self.assertEqual(100, 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(['xa'], env.params['reward_features'])
self.assertEqual(5, env.params['seed'])
#this line is required by Python in order to use multi-processing
if __name__ == '__main__':
# These configuration changes aren't ever required.
# They are simply here to serve as an example.
# These can also be set automatically by creating a .coba file your project root.
CobaContext.cacher.cache_directory = './.coba_cache'
CobaContext.experiment.processes = 2
CobaContext.experiment.chunk_by = 'task'
#First, we define the learners that we want to test
learners = [
RandomLearner(),
EpsilonBanditLearner(),
VowpalEpsilonLearner(),
]
#Next we create the environments we'd like evaluate against
environments = Environments.from_linear_synthetic(
1000, n_action_features=0).shuffle([0, 1, 2, 3])
#We then create and evaluate our experiment from our environments and learners
result = Experiment(environments, learners).evaluate()
#After evaluating can create a quick summary plot to get a sense of how the learners performed
result.plot_learners(err='sd')
#We can also create a plot examining how specific learners did across each shuffle of our environments
result.filter_lrn(full_name="vw").plot_learners(err='sd', each=True)