class Optimizer():
def __init__(self, input_file, N):
self.input_constraints = Constraint(input_file)
#print(self.input_constraints.get_ndim())
#print(self.input_constraints.get_example())
#print(self.input_constraints.get_constraints())
#exam = np.array(list(self.input_constraints.get_example()), dtype=float)
#print(self.input_constraints.apply(exam))
self.x = self.sample(N)
self.get_output()
def sample(self, N):
count = 0
ndim = int(self.input_constraints.get_ndim())
x = np.empty((0, ndim))
while (count < int(N)):
n_samples = int(N) * 10
smpls = lhs(ndim, samples=n_samples)
for i in range(len(smpls)):
if self.input_constraints.apply(smpls[i]) and (count < int(N)):
#print(smpls[i])
x = np.append(x, np.array([smpls[i]]), axis=0)
count += 1
print(count)
print(x.shape)
return x
def get_output(self):
return self.x
class Sampler:
def __init__(self, filename, max_points, max_time):
self.constraint = Constraint(filename)
self.n_dim = self.constraint.n_dim
self.example = self.constraint.example
self.max_points = max_points
self.max_time = max_time
def test_inside(self, x):
if np.any(np.less(x, 0.)):
return False
if np.any(np.greater(x, 1.)):
return False
return self.constraint.apply(x)
def line_search(self, x, direction):
tiny = 1e-6
low, high = 0., np.sqrt(self.n_dim) + tiny
for _ in range(20):
middle = 0.5 * (low + high)
inside = self.test_inside(x + middle * direction)
if inside:
low = middle
else:
high = middle
return low
def random_direction(self):
vector = np.random.normal(size=self.n_dim)
return vector / np.linalg.norm(vector)
def hit_and_run(self):
assert self.test_inside(self.example)
points = [np.array(self.example)]
start_time = time.time()
while len(points) < self.max_points and time.time(
) - start_time < self.max_time:
direction = self.random_direction()
factor = self.line_search(points[-1], direction)
if factor > 0.:
new_point = points[-1] + np.random.uniform(
) * factor * direction
if self.test_inside(new_point):
points.append(new_point)
return points
def sampler(input_file, output_file, n_result):
ct = Constraint(input_file)
ndim = ct.get_ndim()
current_results = []
while len(current_results) < n_result:
res = np.random.random(ndim)
res = (res / sum(res) * sum(ct.get_example())).round(5)
if not ct.apply(res):
continue
current_results.append(res)
print('example:', ct.get_example())
print('res:', res)
with open(output_file, 'a') as f:
for l, el in enumerate(current_results):
string = ' '.join(map(str, el))
# for item in string:
f.write(string)
f.write('\n')
return current_results
def get_constraints(input_file):
"""reads the input file and creates an instance of the Constraint class
"""
# create a new instance of the Constraint class by reading input_file
# catch several errors: file not found, first two lines (number of
# dimensions and starting vector) not formatted correctly, or syntax
# error in the constraints
try:
space = Constraint(input_file)
except FileNotFoundError:
sys.exit("Error: input file not found")
except ValueError:
sys.exit("Error: input file formatted improperly")
except SyntaxError:
sys.exit("Error: syntax error in constraints")
# get the example point and make sure it has the correct dimensionality
example = space.get_example()
if len(example) < space.get_ndim():
sys.exit("Error: example point does not have enough dimensions")
# check and make sure the example point actually satisfies all of the
# constraints. This additionally serves to make sure the constraints
# are specified correctly
try:
check_example = space.apply(example)
except IndexError:
sys.exit("Error: invalid constraints")
except NameError:
sys.exit("Error: invalid constraints")
# if space.apply(example) returned false, then throw an error
if not check_example:
sys.exit("Error: example point is invalid")
return space
def validate_output_file(self, constraints_file_name, points_file_name,
expected):
"""
Validates output file against constraint file to see if required number of points are available.
:param string constraints_file_name: path to a file that contains the constraints
:param string points_file_name: path to a file that contains the vectors to be validated
:param int expected: Number of points that are expected in this file
"""
self.log.info(
f'Running validation: {constraints_file_name} <-> {points_file_name}'
)
# Return if constraints or points files are not found.
try:
constraints = Constraint(constraints_file_name)
except FileNotFoundError:
self.log.critical(
f'Constraints file {constraints_file_name} not found. Skipping validation...'
)
return
try:
points_file = open(points_file_name, 'r')
except FileNotFoundError:
self.log.critical(
f'Points file {points_file_name} not found. Skipping validation...'
)
return
# Counters for number of points that were generated, passed, failed,
generated = 0
passed = 0
failed = 0
for itr, line in enumerate(points_file):
x = [float(_) for _ in line.strip().split(" ")]
# Log how many points have been checked.
if itr != 0 and (itr % pow(10, floor(log10(itr + 1))) == 0):
self.log.debug(f'Checked {itr}')
# Count number of points generated/passed/failed
generated += 1
try:
if constraints.apply(x):
passed += 1
else:
self.log.warning(f'Point outside constraints: {x}')
failed += 1
except:
self.log.error(f'{sys.exc_info()[1]}, point: {x}')
points_file.close()
# Points file summary
self.log.info(f'{generated} generated, {expected} expected')
self.log.info(f'{passed} passed, {failed} failed')
if generated != expected:
self.log.warning(f'Unexpected number of points')
if failed > 0:
self.log.warning(f'{failed} points failed')
if passed == expected or passed == generated:
self.log.info(f'All points are valid')
self.log.info(
f'Validated {constraints_file_name} <-> {points_file_name}')
self.log.info(f'-- Validation complete -- ')
return
class Solution():
def __init__(self, fname, outfile, n_results):
self.con = Constraint(fname)
self.num_x = self.con.n_dim
self.n_results = n_results
self.outfile = outfile
with open(self.outfile, 'w') as f:
f.write(" ".join(map(str, self.con.example)))
f.write("\n")
self.ineq_cons = {
'type': 'ineq',
'fun': lambda x: self.con.eval_con(x.tolist())
}
bound = [0.0, 1.0]
self.bounds = np.asarray([bound] * self.num_x)
# determining the magnitude of the largest initial point
max_initial = max(self.con.example)
if max_initial == 0.0:
self.alpha = 1.0
else:
self.alpha = 1.0 / max_initial
def solve(self):
x0_all = np.random.random_sample(
(self.n_results * 2, self.num_x)) / self.alpha
output = []
cnt = 0
for i in range(len(x0_all)):
x0 = x0_all[i, :]
if self.con.apply(x0.tolist()):
x_feasible = x0
else:
try:
res = self.optimize_feasible(x0)
x_feasible = res.x
if res.status != 0 or any(x_feasible < 0.0) or any(
x_feasible > 1.0) or list(x_feasible) in output:
continue
except:
continue
# if self.con.apply(x_feasible.tolist()):
# print "True"
with open(self.outfile, 'a') as f:
f.write(" ".join(map(str, x_feasible)))
f.write("\n")
output.append(list(x_feasible))
cnt += 1
if cnt == self.n_results:
break
return output
def optimize_feasible(self, x0):
res = optimize.minimize(self.f,
x0,
method='SLSQP',
constraints=[self.ineq_cons],
options={
'ftol': 1e-8,
'disp': False
},
bounds=self.bounds)
return res
def f(self, x):
constr = self.con.eval_con(x.tolist())
constr[np.where(constr > 0)] = 0
return np.sum(np.square(constr))
def constraint(self, x):
constr = self.con.eval_con(x.tolist())
return constr
class SCMC():
"""Use sequentially constrained Monte Carlo method to sample given constrained domains uniformly
Methods
-------
write_ouput:
write the final sample in given path
get_correctness:
return the correctness of the final state
get_history:
return a List, the history all samples in the sequential diffusion process
plot_results:
scatter plot the results along two arbitrary axes
params
------
comp1, int: first axis (h)
comp2, int: second axis (v)
n_iter, int: if want history instead of final state
print_constraints:
print the original constraints
"""
def __init__(self,
input_path,
n_results,
beta_max=10**4,
p_beta=1,
p_rw_step=0,
track_correctness=False,
threshold=.999):
self.input_path = input_path
self.constraints = Constraint(input_path)
constraints_funcs = self.constraints.get_functions()
self.n_dim = self.constraints.get_ndim()
self.n_results = n_results
self.beta_max = beta_max
self.p_beta = p_beta
self.p_rw_step = p_rw_step
#call sampling method scmc
i_sample = init_sample(self.n_dim, self.n_results,
self.constraints.bounds,
self.constraints.get_example())
self.i_samples = []
self.i_samples.append(i_sample)
self.histories = []
self.correctnesses = []
print(len(constraints_funcs))
for i in range(1, len(constraints_funcs) + 1):
history, correctness = scmc(
self.n_dim,
self.n_results,
self.i_samples[i - 1],
constraints_funcs[0:i],
beta_max,
self.constraints.bounds,
p_beta,
p_rw_step,
track_correctness=track_correctness,
given_example=self.constraints.get_example(),
threshold=threshold)
self.i_samples.append(history[-1])
self.histories.append(history)
self.correctnesses.append(correctness)
self.results = self.histories[-1][-1]
# self.history, self.correctness = scmc(self.n_dim, self.n_results, i_sample ,constraints_funcs, beta_max, self.constraints.bounds,
# p_beta,p_rw_step,
# track_correctness=track_correctness,
# given_example = self.constraints.get_example())
# self.results = self.history[-1]
# self.history, self.correctness = scmc(self.n_dim, self.n_results, constraints_funcs, beta_max, self.constraints.bounds,
# p_beta,p_rw_step,
# track_correctness=track_correctness,
# given_example = self.constraints.get_example())
# self.results = self.history[-1]
# self.write_ouput(output_path)
def write_output(self, output_path):
"""
save the sample to a output_path
"""
np.savetxt(output_path, self.results)
def get_correctness(self):
"""
get correctness
"""
correctness = [self.constraints.apply(x) for x in self.results]
correctness = sum(correctness) / self.n_results
return correctness
def get_history(self):
return self.history
def get_results(self):
"""
get the sample
"""
return self.results
def plot_results(self, comp1, comp2, n_iter=None):
"""
Plot the sample in x[comp1],x[comp2]
"""
if n_iter is None:
sample = self.results
else:
sample = self.history[n_iter - 1]
plt.figure()
plt.scatter(sample[:, comp1], sample[:, comp2])
plt.xlim(0, 1)
plt.ylim(0, 1)
def plot_all_axis(self, n_iter=None):
n_dim = self.n_dim
for i in range(n_dim):
for j in range(i + 1, n_dim):
self.plot_results(i, j, n_iter)
plt.xlabel('x_{}'.format(i))
plt.ylabel('x_{}'.format(j))
def print_constraints(self):
print(self.constraints.get_exprs_string())