def make_sigmoid(config):
bench = SigmoidBenchmark()
for k in config.keys():
if k == "action_values":
bench.set_action_values(config[k])
else:
bench.config[k] = config[k]
return bench.get_environment()
def test_init(self):
bench = SigmoidBenchmark()
bench.set_action_values((3, ))
env = bench.get_environment()
wrapped = PolicyProgressWrapper(env, compute_optimal_sigmoid)
self.assertTrue(len(wrapped.policy_progress) == 0)
self.assertTrue(len(wrapped.episode) == 0)
self.assertFalse(wrapped.compute_optimal is None)
def test_render(self, mock_plt):
bench = SigmoidBenchmark()
bench.set_action_values((3, ))
env = bench.get_environment()
env = PolicyProgressWrapper(env, compute_optimal_sigmoid)
for _ in range(2):
done = False
env.reset()
while not done:
_, _, done, _ = env.step(1)
env.render_policy_progress()
self.assertTrue(mock_plt.show.called)
def test_step(self):
bench = SigmoidBenchmark()
bench.set_action_values((3, ))
bench.config.instance_set = [[0, 0], [1, 1], [3, 4], [5, 6]]
env = bench.get_environment()
wrapped = PolicyProgressWrapper(env, compute_optimal_sigmoid)
wrapped.reset()
_, _, done, _ = wrapped.step(1)
self.assertTrue(len(wrapped.episode) == 1)
while not done:
_, _, done, _ = wrapped.step(1)
self.assertTrue(len(wrapped.episode) == 0)
self.assertTrue(len(wrapped.policy_progress) == 1)
return instances
# Helper method to print current set
def print_instance_set(instance_set):
c = 1
for i in instance_set:
print(f"Instance {c}: {i[0]}, {i[1]}")
c += 1
# Make Sigmoid benchmark
bench = SigmoidBenchmark()
bench.set_action_values([3])
# First example: read instances from default instance set path
instances_from_file = bench.get_environment()
print("Instance set read from file")
print_instance_set(instances_from_file.instance_set)
print("\n")
# Second example: Sample instance set before training
instance_set = sample_instance(20)
bench.config.instance_set = instance_set
instances_sampled_beforehand = bench.get_environment()
print("Instance set sampled before env creation")
print_instance_set(instances_sampled_beforehand.instance_set)
print("\n")
def test_action_value_setting(self):
bench = SigmoidBenchmark()
bench.set_action_values([1, 2, 3])
self.assertTrue(bench.config.action_values == [1, 2, 3])
self.assertTrue(bench.config.action_space_args == [6])
self.assertTrue(len(bench.config.observation_space_args[0]) == 10)