def test_compare_maxweight_vs_lp_maxweight_double_reentrant_line_model(
self):
seed = 42
np.random.seed(seed)
initial_state = 50 * np.ones((5, 1))
env = examples.double_reentrant_line_model(
alpha=1,
mu1=4,
mu2=3,
mu3=2,
mu4=3,
mu5=4,
cost_per_buffer=np.array([1, 1, 1, 1, 1])[:, None],
initial_state=initial_state,
capacity=np.ones((5, 1)) * np.inf,
job_conservation_flag=True,
job_gen_seed=seed,
max_episode_length=None)
mw_agent = mw.MaxWeightAgent(env)
mw_lp_agent = mw_lp.MaxWeightLpAgent(env)
states = np.random.randint(50, size=(400, 5))
for s in states:
z_star_mw_lp, _ = mw_lp_agent.max_weight_policy(s[:, None])
z_star_mw = mw_agent.max_weight_policy(s[:, None])
self.assert_both_mw_policies_are_equivalent(
s, env.cost_per_buffer,
env.job_generator.buffer_processing_matrix, z_star_mw_lp,
z_star_mw)
def perform_test_max_weight_binary_policy(state, env, z_star_theory,
method, binary_action):
agent = mw_lp.MaxWeightLpAgent(env,
method=method,
binary_action=binary_action)
z_star, _ = agent.max_weight_policy(state)
equal = 0
for z in z_star_theory:
if np.allclose(z, z_star):
equal = equal + 1
assert equal == 1
def test_compare_maxweight_vs_maxweight_scheduling(self):
seed = 42
np.random.seed(seed)
initial_state = 50 * np.ones((5, 1))
env = examples.double_reentrant_line_model(initial_state=initial_state,
job_gen_seed=seed)
sch_mw_agent = sch_mw.SchedulingMaxWeightAgent(env)
mw_agent = mw_lp.MaxWeightLpAgent(env)
states = np.random.randint(50, size=(400, 5))
for s in states:
z_star_mw, _ = mw_agent.max_weight_policy(s[:, None])
z_star_sch_mw = sch_mw_agent.scheduling_max_weight_policy(s[:,
None])
self.assert_both_mw_policies_are_equivalent(
s, env.cost_per_buffer,
env.job_generator.buffer_processing_matrix, z_star_mw,
z_star_sch_mw)
def test_integration_double_reentrant_line_model(self):
seed = 42
np.random.seed(seed)
initial_state = 50 * np.ones((5, 1))
env = examples.double_reentrant_line_model(
alpha=1,
mu1=4,
mu2=3,
mu3=2,
mu4=3,
mu5=4,
cost_per_buffer=np.array([1, 1, 1, 1, 1])[:, None],
initial_state=initial_state,
capacity=np.ones((5, 1)) * np.inf,
job_conservation_flag=True,
job_gen_seed=seed,
max_episode_length=None)
agent = mw_lp.MaxWeightLpAgent(env)
simulator = ps.SncSimulator(env, agent, discount_factor=0.95)
data = simulator.run(num_simulation_steps=1000)
assert np.all(data['state'][-1] < initial_state)
def test_compare_maxweight_vs_lp_maxweight_simple_routing_model(self):
seed = 42
np.random.seed(seed)
env = examples.simple_routing_model(alpha_r=0.2,
mu1=0.13,
mu2=0.07,
mu_r=0.2,
cost_per_buffer=np.ones((3, 1)),
initial_state=(1, 1, 1),
capacity=np.ones((3, 1)) * np.inf,
job_conservation_flag=True,
job_gen_seed=seed,
max_episode_length=None)
mw_agent = mw.MaxWeightAgent(env)
mw_lp_agent = mw_lp.MaxWeightLpAgent(env)
states = np.random.randint(50, size=(400, 3))
for s in states:
z_star_mw_lp, _ = mw_lp_agent.max_weight_policy(s[:, None])
z_star_mw = mw_agent.max_weight_policy(s[:, None])
self.assert_both_mw_policies_are_equivalent(
s, env.cost_per_buffer,
env.job_generator.buffer_processing_matrix, z_star_mw_lp,
z_star_mw)
def perform_test_max_weight_policy(state, env, z_star_theory):
agent = mw_lp.MaxWeightLpAgent(env)
z_star, _ = agent.max_weight_policy(state)
np.testing.assert_almost_equal(z_star, z_star_theory)