omega2[idx] = omega[1:]
l = range(i * N, (i + 1) * N)
# random phase shift on [0, 2pi) following Saltelli et al. Technometrics 1999
phi = 2 * math.pi * np.random.rand()
for j in range(D):
g = 0.5 + (1 / math.pi) * np.arcsin(np.sin(omega2[j] * s + phi))
X[l, j] = g
scale_samples(X, pf['bounds'])
return X
if __name__ == "__main__":
parser = common_args.create()
parser.add_argument('-M',
type=int,
required=False,
default=4,
help='M coefficient, default 4')
args = parser.parse_args()
np.random.seed(args.seed)
param_values = sample(args.samples, args.paramfile, M=args.M)
np.savetxt(args.output,
param_values,
delimiter=args.delimiter,
fmt='%.' + str(args.precision) + 'e')
def calc_vi(base, perturbed, x_delta):
# v_i sensitivity measure following Sobol and Kucherenko (2009)
# For comparison, Morris mu* < sqrt(v_i)
dfdx = (perturbed-base)/x_delta
dfdx2 = dfdx**2
return np.mean(dfdx2), np.std(dfdx2)
def calc_dgsm(base, perturbed, x_delta, bounds, num_resamples, conf_level):
# v_i sensitivity measure following Sobol and Kucherenko (2009)
# For comparison, total order S_tot <= dgsm
D = np.var(base)
vi, _ = calc_vi(base, perturbed, x_delta)
dgsm = vi*(bounds[1]-bounds[0])**2/(D*np.pi**2)
s = np.empty(num_resamples)
for i in xrange(num_resamples):
r = np.random.randint(len(base), size=len(base))
s[i], _ = calc_vi(base[r], perturbed[r], x_delta[r])
return dgsm, norm.ppf(0.5 + conf_level/2) * s.std(ddof=1)
if __name__ == "__main__":
parser = common_args.create()
parser.add_argument('-X', '--model-input-file', type=str, required=True, default=None, help='Model input file')
parser.add_argument('-r', '--resamples', type=int, required=False, default=1000, help='Number of bootstrap resamples for Sobol confidence intervals')
args = parser.parse_args()
analyze(args.paramfile, args.model_input_file, args.model_output_file, args.column,
num_resamples = args.resamples, delim = args.delimiter, print_to_console=True)
