def test_evaluate_mrp(self):
v = evaluate_mrp_result(self.finite_flip_flop, gamma=0.99)
self.assertEqual(len(v), 2)
for s in v:
self.assertLess(abs(v[s] - 170), 0.1)
def test_compare_to_backward_induction(self):
finite_horizon = finite_horizon_MRP(self.finite_flip_flop, 10)
v = evaluate_mrp_result(finite_horizon, gamma=1)
self.assertEqual(len(v), 20)
finite_v =\
list(evaluate(unwrap_finite_horizon_MRP(finite_horizon), gamma=1))
for time in range(0, 10):
self.assertAlmostEqual(v[WithTime(state=True, time=time)],
finite_v[time][True])
self.assertAlmostEqual(v[WithTime(state=False, time=time)],
finite_v[time][False])
print("---------------")
implied_mrp.display_reward_function()
print()
print("Implied MRP Value Function")
print("--------------")
implied_mrp.display_value_function(gamma=user_gamma)
print()
from rl.dynamic_programming import evaluate_mrp_result
from rl.dynamic_programming import policy_iteration_result
from rl.dynamic_programming import value_iteration_result
print("Implied MRP Policy Evaluation Value Function")
print("--------------")
pprint(evaluate_mrp_result(implied_mrp, gamma=user_gamma))
print()
print("MDP Policy Iteration Optimal Value Function and Optimal Policy")
print("--------------")
opt_vf_pi, opt_policy_pi = policy_iteration_result(fe_mdp,
gamma=user_gamma)
pprint(opt_vf_pi)
print(opt_policy_pi)
print()
print("MDP Value Iteration Optimal Value Function and Optimal Policy")
print("--------------")
opt_vf_vi, opt_policy_vi = value_iteration_result(fe_mdp, gamma=user_gamma)
pprint(opt_vf_vi)
print(opt_policy_vi)