def invert_bootstrap(self):
"""
It is used to evaluate the mean, standard deviation, and variance
of new parameters
:return:
"""
A_bootstrap = []
b_bootstrap = []
n_subset = int(const.BOOTSTRAP_SUBSET_RATIO * self.nwins)
for i in range(self.config.bootstrap_repeat):
random_array = util.gen_random_array(
self.nwins, sample_number=n_subset)
A = util.sum_matrix(random_array * self.weight_array, self.A1_all)
b = util.sum_matrix(random_array * self.weight_array, self.b1_all)
A_bootstrap.append(A)
b_bootstrap.append(b)
# inversion of each subset
new_par_array = np.zeros((self.config.bootstrap_repeat, const.NPARMAX))
for i in range(self.config.bootstrap_repeat):
new_par = self.invert_solver(A_bootstrap[i], b_bootstrap[i])
new_par_array[i, :] = new_par
# statistical analysis
self.par_mean = np.mean(new_par_array, axis=0)
self.par_std = np.std(new_par_array, axis=0)
self.par_var = np.var(new_par_array, axis=0)
for _ii in range(self.par_mean.shape[0]):
if self.par_mean[_ii] != 0:
# in case of 0 value
self.std_over_mean[_ii] = \
np.abs(self.par_std[_ii] / self.par_mean[_ii])
else:
self.std_over_mean[_ii] = 0.
def invert_cmt(self):
"""
ensemble all measurements together to form Matrix A and vector
b to solve the A * (dm) = b
A is the Hessian Matrix and b is the misfit
:return:
"""
logger.info("*"*15)
logger.info("CMT Inversion")
logger.info("*"*15)
# ensemble A and b
A = util.sum_matrix(self.weight_array, self.A1_all)
b = util.sum_matrix(self.weight_array, self.b1_all)
logger.info("Inversion Matrix A is as follows:")
logger.info("\n%s" % ('\n'.join(map(self._float_array_to_str, A))))
logger.info("Condition number of A: %10.2f" % (np.linalg.cond(A)))
logger.info("RHS vector b is as follows:")
logger.info("[%s]" % (self._float_array_to_str(b)))
# source inversion
self.new_cmt_par = self.invert_solver(A, b, print_mode=True)
self.convert_new_cmt_par()
