# ----------------------------------------------------------------------------
# First filter NaN values in case run did not finish
filt_models, filt_energy, niter = pp.NaN_filter(models, energy)
del models, energy
# ---> Prefilter models according to parameter space regular subgriding
# !!! Care must be taken that we do this parameter log
nruns, popsize, nparam, niter = pp.filt_models.shape
print "number of models to filter: ", "{:e}".format(nruns * popsize * niter)
# --- > regrid parameter space
delta_m = 1e-3
m_grid, f_grid, rgrid_error = pp.regrid(filt_models,
filt_energy,
delta_m,
center=True)
print "number of particles in subgrid", "{:e}".format(m_grid.shape[0])
print "Error in energy minimum after subgrid :", np.min(f_grid) - f_gbest
# ---> save m_grid, f_grid, r_grid_error, m_gbest,
# f_best, delta_m
if save_netcdf:
if os.path.isfile(outfile):
os.remove(outfile)
print "remove ", outfile
nc = Dataset(outfile, "w", format='NETCDF4')
# dimensions: name, size
nc.createDimension('nruns', m_grid.shape[0])
nc.createDimension('popsize', m_grid.shape[1])
# ----------------------------------------------------------------------------
# First filter NaN values in case run did not finish
filt_models, filt_energy, niter = pp.NaN_filter(models[:irun, : , :, :],
energy[:irun, : , :])
# ---> Prefilter models according to parameter space regular subgriding
# Care must be taken that we do this parameter log
nruns, popsize, nparam, niter = filt_models.shape
print "number of models to filter: ", "{:e}".format(nruns * popsize * niter)
print''
threshold = np.max(filt_energy) # np.median(filt_energy)
m_near, f_near = pp.value_filter(filt_models, filt_energy, threshold)
# --- > regrid parameter space
delta_m = 1e-3
m_grid, f_grid, rgrid_error = pp.regrid(m_near, f_near, delta_m, center=True)
del m_near, f_near
print "number of particles in subgrid", "{:e}".format(m_grid.shape[0])
print "Error in energy minimum after subgrid :", np.min(f_grid) - f_gbest
print''
# ---> Xi2 weighted mean model, in log and physical space
m_weight = pp.weighted_mean(m_grid, f_grid, ndata, kappa=1, rms=True, log=True)
# ---> Xi2 weighted STD model, in log and physical space
std_weight = pp.weighted_std(m_weight, m_grid, f_grid, ndata, kappa=1, rms=True, log=True)
# ---- marginal laws using RMS
indx = indx + 1
pdf_m[indx, :, :], n_bin, x_bin = pp.marginal_law(m_grid, f_grid, logrhosynth, ndata,
n_inter=n_inter,lower=lower, upper=upper,
in_far_p = upper - 2 * delta_mgrid
in_close_m = -in_close_p
in_m = -in_p
in_far_m = -in_far_p
models = np.array([
out_big_m, out_m, out_small_m, in_close_m, in_m, in_far_m, in_close_p,
in_p, in_far_p, out_big_p, out_p, out_small_p
])
f_near = models * 5
# ---> regrid ok !
m_grid, f_grid, rgrid_error = pp.regrid(models,
f_near,
delta_m=delta_mgrid,
center=True)
m_non_unique = (models + delta_mgrid / 2) // delta_mgrid * delta_mgrid
# ---> marginal_law ?
# is epsilon relevant ?
nparam = 1
n_inter = 40
n_bin = np.empty(shape=(n_inter))
pdf_m = np.empty(shape=(n_inter))
x_bin = np.empty(shape=(n_inter))
L = upper - lower
eps = 1e-3
p_inter = np.arange(n_inter + 1.) * L / n_inter - L * 0.5