def test_compile_4():
mdp = MDP()
t1 = Transition("s0", "a0", "s1", 1, 0)
mdp.add_transition(t1)
with pytest.raises(ValueError):
mdp.ensure_compiled()
mdp.add_terminal_state("s1")
mdp.ensure_compiled()
def test_compile_5():
mdp = MDP()
t1 = Transition("s0", "a0", "s1", 1, 0)
mdp.add_transition(t1)
mdp.add_terminal_state("s1")
mdp.ensure_compiled()
assert mdp.compiled
assert type(mdp.states) is tuple
assert type(mdp.actions) is tuple
assert type(mdp.terminal_states) is tuple
mdp._decompile()
assert not mdp.compiled
assert type(mdp.states) is set
assert type(mdp.actions) is set
assert type(mdp.terminal_states) is set
def test_set_terminal():
mdp = MDP()
t1 = Transition("s0", "a0", "s1", 0, 0)
mdp.add_transition(t1)
mdp.add_terminal_state("s0")
assert len(mdp.states) == 2
assert len(mdp.actions) == 1
assert len(mdp.terminal_states) == 1
assert "s0" in mdp.terminal_states
mdp.add_terminal_state("s4")
assert len(mdp.states) == 3
assert len(mdp.terminal_states)
assert "s4" in mdp.terminal_states
assert "s4" in mdp.states
:param theta (float, optional): stop threshold, defaults to 1e-6
:return (Tuple[np.ndarray of float with dim (num of states, num of actions),
np.ndarray of float with dim (num of states)]):
Tuple of calculated policy and value function
"""
self.mdp.ensure_compiled()
self.theta = theta
return self._policy_improvement()
if __name__ == "__main__":
mdp = MDP()
mdp.add_transition(
# start action end prob reward
Transition("high", "wait", "high", 1, 2),
Transition("high", "search", "high", 0.8, 5),
Transition("high", "search", "low", 0.2, 5),
Transition("high", "recharge", "high", 1, 0),
Transition("low", "recharge", "high", 1, 0),
Transition("low", "wait", "low", 1, 2),
Transition("low", "search", "high", 0.6, -3),
Transition("low", "search", "low", 0.4, 5),
)
solver = ValueIteration(mdp, 0.9)
policy, valuefunc = solver.solve()
print("---Value Iteration---")
print("Policy:")
print(solver.decode_policy(policy))
print("Value Function")