def test_qr_complex(self):
shapes = [(20, 15), (15, 20), (20, 20)]
for m, n in shapes:
np.random.seed(1) # always get the same random
a = np.random.rand(m, n).astype(np.complex_) + 1j * np.random.rand(m, n)
q, r = qr(a, reduced=False)
self.assertAlmostEqual(0, np.absolute(a - q @ r).sum()) # val.1
if m == n:
self.assertAlmostEqual(q.T.conj().all(), np.linalg.inv(q).all()) # val.6
self.assertAlmostEqual(1, norm(np.linalg.det(q))) # val. 3
self.assertTrue(np.allclose(q.T.conj() @ q, q @ q.T.conj())) # val.4
self.assertTrue(np.allclose(np.identity(q.shape[0]), q @ q.T.conj())) # val.5
self.assertTrue(np.allclose(np.triu(r), r)) # val.7
q, r = qr(a)
mgs_q, mgs_r = mgs(a)
self.assertTrue(np.allclose(np.absolute(mgs_q), np.absolute(q))) # val.2
self.assertTrue(np.allclose(np.absolute(mgs_r), np.absolute(r))) # val.2
a = np.array([[5. + 5.j, -5.j, 2.],
[-4.j, 8. + 8.j, -6.],
[0., -6., 10.]])
b = np.array([3., 5. + 4.j, -2.]).reshape(-1, 1)
x_expected = np.array([[0.6149065 + 0.1621229j],
[0.7862721 + 0.0385111j],
[0.2717633 + 0.0231067j]])
x_actual = solve(a, b)
self.assertAlmostEqual(x_expected.all(), x_actual.all()) # val.8
def fit_to_sphere(point, radius=1):
# fit one 3-dim vector to a sphere
# applies to numpy array with 3d rows
# freaks out when point is 0
try:
fitted = radius * point / norm(point)
except:
fitted = 0
return fitted
def extract_lengths(data):
lengths = dict()
for seg in data_model.KIN_TREE:
# for now, stupidly average over whole sample
try:
lengths[seg] = np.mean(norm(np.array(data[seg[1]]) - np.array(data[seg[0]])))
except KeyError:
pass
return lengths
def test_norm(self):
v_r = np.array([[5.], [4.], [1.]])
self.assertEqual(np.sqrt(5.**2 + 4.**2 + 1.), norm(v_r))
v_c = v_r + 1j * np.array([[1.], [0.], [3.]])
self.assertEqual(np.sqrt(5.**2 + 1. + 4.**2 + 1. + 3**2), norm(v_c))
def run(self):
while True:
# Get the work from the queue and expand the tuple
current_env_variables, current_iteration, buffer_env_predicate_pairs = self.queue.get(
)
self.logger.info(
"Env variables for binary search: {}, Iteration: {}".format(
current_env_variables.get_params_string(),
current_iteration))
t_env_variables = copy.deepcopy(self.init_env_variables)
f_env_variables = copy.deepcopy(current_env_variables)
binary_search_counter = 0
max_binary_search_iterations = 20
search_suffix = "binary_search_"
if self.param_names:
search_suffix += self.param_names_string + "_"
search_suffix += str(current_iteration) + "_" + str(
binary_search_counter)
dist = norm(
env_vars_1=f_env_variables,
env_vars_2=t_env_variables) / norm(env_vars_1=t_env_variables)
execution_times = []
regression_times = []
env_predicate_pairs = []
while dist > self.binary_search_epsilon:
new_env_variables = get_binary_search_candidate(
t_env_variables=t_env_variables,
f_env_variables=f_env_variables,
algo_name=self.algo_name,
env_name=self.env_name,
param_names=self.param_names,
discrete_action_space=self.discrete_action_space,
buffer_env_predicate_pairs=buffer_env_predicate_pairs,
)
self.logger.debug("New env after binary search: {}".format(
new_env_variables.get_params_string()))
env_predicate_pair, execution_time, regression_time = self.runner.execute_train(
current_iteration=current_iteration,
search_suffix=search_suffix,
current_env_variables=new_env_variables,
_start_time=self.start_time,
)
execution_times.append(execution_time)
regression_times.append(regression_time)
env_predicate_pairs.append(env_predicate_pair)
if env_predicate_pair.is_predicate():
self.logger.debug("New t_env found: {}".format(
env_predicate_pair.get_env_variables(
).get_params_string()))
t_env_variables = copy.deepcopy(
env_predicate_pair.get_env_variables())
else:
self.logger.debug("New f_env found: {}".format(
env_predicate_pair.get_env_variables(
).get_params_string()))
f_env_variables = copy.deepcopy(
env_predicate_pair.get_env_variables())
dist = norm(env_vars_1=f_env_variables,
env_vars_2=t_env_variables) / norm(
env_vars_1=t_env_variables)
if binary_search_counter == max_binary_search_iterations:
break
binary_search_counter += 1
search_suffix = "binary_search_"
if self.param_names:
search_suffix += self.param_names_string + "_"
search_suffix += str(current_iteration) + "_" + str(
binary_search_counter)
self.logger.info("dist {} <= binary_search_epsilon {}".format(
dist, self.binary_search_epsilon))
self.queue_result.put_nowait((
t_env_variables,
f_env_variables,
env_predicate_pairs,
max_binary_search_iterations,
binary_search_counter,
current_iteration,
execution_times,
regression_times,
))
self.queue.task_done()
