def check_forces_reproducibility(file_name, n_iter=500):
# minimizer = 'lbfgs'
minimizer = 'gsl'
params = optimize.minimize.Parameters(minimizer)
params['cache_ytilde_transposed'] = True
params['use_c_functions'] = True
# params['algorithm'] = "lbfgs"
params['algorithm'] = "bfgs2"
params['verbose'] = False
[forces_init, w0, y, yTilde, YTilde, theta] = fio.load(file_name,
hdf5_keys=["forces_init", "w0", "y", "yTilde", "YTilde", "theta"])
fmin_list = []
forces_sum_list = []
for i in range(n_iter):
wopt, yopt, forces, fmin_initial, fmin_final, chiSqr, S = \
optimize.forces.find_optimum(forces_init, w0, y, yTilde, YTilde, theta, params)
fmin_list.append(fmin_final)
forces_sum_list.append(np.sum(forces))
fmin_diff_list = []
for i in range(1, n_iter):
diff = optimize.util.compute_relative_difference_for_values(fmin_list[0], fmin_list[i])
fmin_diff_list.append(diff)
forces_diff_list = []
for i in range(1, n_iter):
diff = optimize.util.compute_relative_difference_for_values(forces_sum_list[0], forces_sum_list[i])
forces_diff_list.append(diff)
assert(np.sum(fmin_diff_list) < tol)
assert(np.sum(forces_diff_list) < tol)
def test_fileio_forces():
if sys.version_info >= (3, ):
return
for filename_pkl in filenames_forces:
hdf5_file = tempfile.NamedTemporaryFile(mode='w',
suffix=".h5",
delete=False)
hdf5_file.close()
print(filename_pkl, hdf5_file.name)
# perform file conversion
keys = ["forces_init", "w0", "y", "yTilde", "YTilde", "theta"]
fio.convert_to_hdf5(filename_pkl, hdf5_file.name, keys)
# load pickle
with open(filename_pkl, 'rb') as ifile:
x = pickle.load(ifile)
# load hdf5
y = fio.load(hdf5_file.name, hdf5_keys=keys)
os.unlink(hdf5_file.name)
# compare
assert (len(x) == len(y))
for i, elem_x in enumerate(x):
elem_y = y[i]
assert (np.array_equal(elem_x, elem_y))
def test_dump_load_list_unlabeled(temp_hdf5_file_name):
filename = temp_hdf5_file_name
data = [1, 2, 3, 5.0]
fio.dump(filename, data)
check = fio.load(filename)
for i, item in enumerate(check):
assert(item == data[i])
os.remove(filename)
def test_dump_load_list_labeled(temp_hdf5_file_name):
filename = temp_hdf5_file_name
data = [1, 2, 3, 5.0]
keys = ["one", "two", "three", "five.zero"]
fio.dump(filename, data, hdf5_keys=keys)
check = fio.load(filename, hdf5_keys=keys)
for i, item in enumerate(check):
assert(item == data[i])
os.remove(filename)
def run_func(use_c=True):
[GInit, G, y, yTilde, YTilde, w0, theta] = fio.load(
filename,
hdf5_keys=["GInit", "G", "y", "yTilde", "YTilde", "w0", "theta"])
g = GInit.copy()
gPrime = np.asarray(g[:].T)[0]
log_posterior = optimize.log_weights.bioen_log_posterior(gPrime,
g,
G,
yTilde,
YTilde,
theta,
use_c=use_c)
return log_posterior
def run_test_error_opt_logw(file_name=filenames[0],
library='scipy/py',
caching=False):
print("=" * 80)
if (create_reference_values):
os.environ["OMP_NUM_THREADS"] = "1"
if "OMP_NUM_THREADS" in os.environ:
print("OPENMP NUM. THREADS = ", os.environ["OMP_NUM_THREADS"])
exp = available_tests()
# load exp. data from file
[GInit, G, y, yTilde, YTilde, w0, theta] = fio.load(
file_name,
hdf5_keys=["GInit", "G", "y", "yTilde", "YTilde", "w0", "theta"])
# run all available optimizations
for minimizer in exp:
for algorithm in exp[minimizer]:
use_c_functions = True
params = optimize.minimize.Parameters(minimizer)
params['cache_ytilde_transposed'] = caching
params['use_c_functions'] = use_c_functions
params['algorithm'] = algorithm
params['verbose'] = verbose
if params['minimizer'] == "gsl":
# Force an error by defining a wrong parameter
#params['params']['step_size'] = -1.00001
params['algorithm'] = "TEST_INVALID"
elif params['minimizer'] == "lbfgs":
# Force an error by defining a wrong parameter
params['params']['delta'] = -1.
if verbose:
print("-" * 80)
print(params)
with pytest.raises(RuntimeError) as excinfo:
# run optimization
wopt, yopt, gopt, fmin_ini, fmin_fin = \
optimize.log_weights.find_optimum(GInit, G, y, yTilde, YTilde, theta, params)
print(excinfo.value)
assert ('return code' in str(excinfo.value))
def test_dump_load_dict(temp_hdf5_file_name):
filename = temp_hdf5_file_name
mynested = {"var1": "a_string", "var2": 32, "var3": [2, 3, 4]}
mydict = {"label": "value", "nested": mynested}
fio.dump(filename, mydict)
mydict = fio.load(filename, hdf5_deep_mode=True)
assert (mydict["label"] == 'value')
assert (mydict["nested"]["var1"] == "a_string")
assert (mydict["nested"]["var2"] == 32)
assert (mydict["nested"]["var3"][0] == 2)
assert (mydict["nested"]["var3"][1] == 3)
assert (mydict["nested"]["var3"][2] == 4)
os.remove(filename)
def run_test_error_forces(file_name=filenames[0], caching=False):
print("=" * 80)
if "OMP_NUM_THREADS" in os.environ:
print(" OPENMP NUM. THREADS = ", os.environ["OMP_NUM_THREADS"])
exp = available_tests()
[forces_init, w0, y, yTilde, YTilde, theta] = fio.load(
file_name,
hdf5_keys=["forces_init", "w0", "y", "yTilde", "YTilde", "theta"])
for minimizer in exp:
for algorithm in exp[minimizer]:
params = optimize.minimize.Parameters(minimizer)
params['cache_ytilde_transposed'] = caching
params['use_c_functions'] = True
params['algorithm'] = algorithm
params['verbose'] = True
if params['minimizer'] == "gsl":
# Force an error by defining a wrong parameter
params['algorithm'] = "TEST_INVALID"
#params['params']['step_size'] = -1.00001
# print (params['params'])
elif params['minimizer'] == "lbfgs":
# Force an error by defining a wrong parameter
params['params']['delta'] = -1.
print("-" * 80)
print("", params)
with pytest.raises(RuntimeError) as excinfo:
wopt, yopt, forces, fmin_ini, fmin_fin, chiSqr, S = \
optimize.forces.find_optimum(forces_init, w0, y, yTilde, YTilde, theta, params)
print(excinfo.value)
assert ('return code' in str(excinfo.value))
def run_grad_forces(use_c=True):
[forces_init, w0, y, yTilde, YTilde, theta] = fio.load(filename,
hdf5_keys=["forces_init", "w0", "y", "yTilde", "YTilde", "theta"])
fprime = optimize.forces.grad_bioen_log_posterior(forces_init, w0, y, yTilde, YTilde, theta, use_c)
return fprime
def run_test_optimum_logw(file_name=filenames[0], library='scipy/py', caching=False):
print("=" * 80)
if (create_reference_values):
os.environ["OMP_NUM_THREADS"] = "1"
if "OMP_NUM_THREADS" in os.environ:
print("OPENMP NUM. THREADS = ", os.environ["OMP_NUM_THREADS"])
exp = available_tests()
# Run the optimizer for all the available tests
for minimizer in exp:
for algorithm in exp[minimizer]:
[GInit, G, y, yTilde, YTilde, w0, theta] = fio.load(file_name,
hdf5_keys=["GInit", "G", "y", "yTilde", "YTilde", "w0", "theta"])
minimizer_tag = minimizer
use_c_functions = True
if (minimizer == 'scipy_py'):
minimizer_tag = 'scipy'
use_c_functions = False
if (minimizer == 'scipy_c'):
minimizer_tag = 'scipy'
# get default parameter's configuration for a minimizer
params = optimize.minimize.Parameters(minimizer_tag)
params['cache_ytilde_transposed'] = caching
params['use_c_functions'] = use_c_functions
params['algorithm'] = algorithm
params['verbose'] = verbose
if verbose:
print("-" * 80)
print(params)
# run optimization
wopt, yopt, gopt, fmin_ini, fmin_fin = \
optimize.log_weights.find_optimum(GInit, G, y, yTilde, YTilde, theta, params)
# store the results in the structure
exp[minimizer][algorithm]['wopt'] = wopt
exp[minimizer][algorithm]['yopt'] = yopt
exp[minimizer][algorithm]['gopt'] = gopt
exp[minimizer][algorithm]['fmin_ini'] = fmin_ini
exp[minimizer][algorithm]['fmin_fin'] = fmin_fin
if (create_reference_values):
print("-" * 80)
for minimizer in exp:
for algorithm in exp[minimizer]:
fmin_fin = exp[minimizer][algorithm]['fmin_fin']
print(" === CREATING REFERENCE VALUES ===")
ref_file_name = os.path.splitext(file_name)[0] + ".ref.h5"
fio.dump(ref_file_name, fmin_min, "reference")
print(" [%8s][%4s] -- fmin: %.16f --> %s" % (minimizer, algorithm, fmin_fin, ref_file_name))
print("=" * 34, " END TEST ", "=" * 34)
print("%" * 80)
else:
ref_file_name = os.path.splitext(file_name)[0] + ".ref.h5"
available_reference = False
if (os.path.isfile(ref_file_name)):
available_reference = True
x = fio.load(ref_file_name, hdf5_deep_mode=True)
fmin_reference = x["reference"]
# print results
print("=" * 80)
print(" [LOGW] CHECK RESULTS for file [ %s ] -- caching(%s)" % (file_name, caching))
if (available_reference):
print("-" * 80)
print(" === REFERENCE EVALUATION ===")
for minimizer in exp:
for algorithm in exp[minimizer]:
fmin_fin = exp[minimizer][algorithm]['fmin_fin']
eval_diff = optimize.util.compute_relative_difference_for_values(fmin_fin, fmin_reference)
print(" [%8s][%16s] -- fmin: %.16f -- fmin_reference : %.16f -- diff(tol=%1.0e) = %.16f" % (
minimizer, algorithm, fmin_fin, fmin_reference, tol_min, eval_diff))
for minimizer in exp:
for algorithm in exp[minimizer]:
fmin_fin = exp[minimizer][algorithm]['fmin_fin']
eval_diff = optimize.util.compute_relative_difference_for_values(fmin_fin, fmin_reference)
assert(np.all(eval_diff < tol_min))
else:
print("-" * 80)
print(" === REFERENCE EVALUATION === ")
print(" [Reference not found] To re-generate reference values enable create_reference_values=True and relaunch")
assert(False)
print("-" * 80)
print(" === RETURNED GRADIENT EVALUATION ===")
# re-evaluation of minimum for the returned vector
[GInit, G, y, yTilde, YTilde, w0, theta] = fio.load(file_name,
hdf5_keys=["GInit", "G", "y", "yTilde", "YTilde", "w0", "theta"])
for minimizer in exp:
for algorithm in exp[minimizer]:
gopt = exp[minimizer][algorithm]['gopt']
exp[minimizer][algorithm]['re_fmin'] = \
optimize.log_weights.bioen_log_posterior(gopt, GInit, G, yTilde, YTilde, theta, use_c=True)
# print differences
for minimizer in exp:
for algorithm in exp[minimizer]:
re_fmin = exp[minimizer][algorithm]['re_fmin']
fmin_fin = exp[minimizer][algorithm]['fmin_fin']
eval_diff = optimize.util.compute_relative_difference_for_values(fmin_fin, re_fmin)
print(" [%8s][%16s] -- fmin: %.16f -- fmin_for_gradient: %.16f -- diff(tol=%1.0e) = %.16f" % (
minimizer, algorithm, fmin_fin, re_fmin, tol, eval_diff))
print("=" * 34, " END TEST ", "=" * 34)
print("%" * 80)
# validate differences
for minimizer in exp:
for algorithm in exp[minimizer]:
re_fmin = exp[minimizer][algorithm]['re_fmin']
fmin_fin = exp[minimizer][algorithm]['fmin_fin']
eval_diff = optimize.util.compute_relative_difference_for_values(fmin_fin, re_fmin)
assert(np.all(eval_diff < tol))
return
