def test_FSARSA_0_run_learning_episode():
environment = FakeContinuousEnvironment()
environment.approximate_with(FuzzyApproximator, fuzzy_sets=fuzzy_sets)
algorithm = fuzzy.FSARSA(environment)
assert sum([algorithm.q[i].sum() for i in algorithm.actions]) == 0.0
env = algorithm.run_learning_episode()
assert isinstance(env, Environment)
assert sum([algorithm.q[i].sum() for i in algorithm.actions]) != 0.0
def test_AHC_lambda_continuous_environment_run_learning_episode():
environment = FakeContinuousEnvironment()
environment.approximate_with(TableApproximator)
algorithm = classic.AHC(environment, lambd=0.2)
assert sum([algorithm.mi[:, i].sum() for i in algorithm.actions]) == 0.0
env = algorithm.run_learning_episode()
assert isinstance(env, Environment)
assert sum([algorithm.mi[:, i].sum() for i in algorithm.actions]) != 0.0
def test_Q_SARSA_R_continuous_environment_run_learning_episode(algorithm, lambd):
environment = FakeContinuousEnvironment()
environment.approximate_with(TableApproximator)
algorithm = algorithm(environment, lambd=lambd)
assert sum([algorithm.Q[:, i].sum() for i in algorithm.actions]) == 0.0
env = algorithm.run_learning_episode()
assert isinstance(env, Environment)
assert sum([algorithm.Q[:, i].sum() for i in algorithm.actions]) != 0.0
def test_CMACQ_CMACSARSA_CMACR_lambda_run_learning_episode(algorithm, lambd):
environment = FakeContinuousEnvironment()
environment.approximate_with(CMACApproximator)
algorithm = algorithm(environment, lambd=lambd)
for l in range(algorithm.n_layers):
assert sum([algorithm.q[l][:, i].sum() for i in algorithm.actions]) == 0.0
env = algorithm.run_learning_episode()
assert isinstance(env, Environment)
for l in range(algorithm.n_layers):
assert sum([algorithm.q[l][:, i].sum() for i in algorithm.actions]) != 0.0
def test_CMACAHC_0_run_learning_episode():
environment = FakeContinuousEnvironment()
environment.approximate_with(CMACApproximator)
algorithm = cmac.CMACAHC(environment)
for l in range(algorithm.n_layers):
assert sum([algorithm.mi[l][:, i].sum() for i in algorithm.actions]) == 0.0
env = algorithm.run_learning_episode()
assert isinstance(env, Environment)
for l in range(algorithm.n_layers):
assert sum([algorithm.mi[l][:, i].sum() for i in algorithm.actions]) != 0.0
def test_Q_SARSA_R_continuous_environment_get_greedy_actions(algorithm):
environment = FakeContinuousEnvironment()
environment.approximate_with(TableApproximator)
algorithm = algorithm(environment)
environment_state = 2
greedy_actions = algorithm.get_greedy_actions(environment_state)
assert np.array_equal(greedy_actions, algorithm.actions)
for i in range(algorithm.Q.shape[0]):
algorithm.Q[i][0] = 1
greedy_actions = algorithm.get_greedy_actions(environment_state)
assert np.array_equal(greedy_actions, algorithm.actions[:1])
def test_CMACAHC_get_greedy_actions():
environment = FakeContinuousEnvironment()
environment.approximate_with(CMACApproximator)
algorithm = cmac.CMACAHC(environment)
environment_state = (0, 0)
greedy_actions = algorithm.get_greedy_actions(environment_state)
assert np.array_equal(greedy_actions, algorithm.actions)
for l in range(algorithm.n_layers):
for i in range(algorithm.mi[l].shape[0]):
algorithm.mi[l][i][0] = 1
greedy_actions = algorithm.get_greedy_actions(environment_state)
assert np.array_equal(greedy_actions, algorithm.actions[:1])
def test_CMACQ_CMACSARSA_CMACR_get_greedy_actions(algorithm, lambd):
environment = FakeContinuousEnvironment()
environment.approximate_with(CMACApproximator)
algorithm = algorithm(environment, lmbd=lambd)
environment_state = (0, 0)
greedy_actions = algorithm.get_greedy_actions(environment_state)
assert np.array_equal(greedy_actions, algorithm.actions)
for l in range(algorithm.n_layers):
for i in range(algorithm.q[l].shape[0]):
algorithm.q[l][i][0] = 1
greedy_actions = algorithm.get_greedy_actions(environment_state)
assert np.array_equal(greedy_actions, algorithm.actions[:1])
def test_FQ_get_greedy_actions():
environment = FakeContinuousEnvironment()
environment.approximate_with(FuzzyApproximator, fuzzy_sets=fuzzy_sets)
algorithm = fuzzy.FQ(environment)
# environment state approximed for variables = (-0.5, -0.5)
environment_state = [np.array([0.5, 0.5, 0.0]), np.array([1.0, 0.0])]
greedy_actions = algorithm.get_greedy_actions(environment_state)
# all of possible actions because all q ale zeros
assert list(greedy_actions) == algorithm.actions
# now second actions q is modified to be greedy
algorithm.q[1][1][0] = 1.0
greedy_actions = algorithm.get_greedy_actions(environment_state)
assert greedy_actions == algorithm.actions[1]
def test_state():
env = FakeContinuousEnvironment(101)
assert env.state == (0.0, 1.0)
env.approximate_with(FakeApproximator)
assert env.state == 1
def test_states():
env = FakeContinuousEnvironment(101)
with pytest.raises(AttributeError):
env.states
env.approximate_with(FakeApproximator)
assert env.states == 2
def test_FQ_get_greedy_actions_invalid_environment_state_given():
environment = FakeContinuousEnvironment()
environment.approximate_with(FuzzyApproximator, fuzzy_sets=fuzzy_sets)
algorithm = fuzzy.FQ(environment)
with pytest.raises(ValueError):
algorithm.get_greedy_actions(np.ones((2,2)))