def array_to_string(array_to_write,dim):
if dim == 0:
string = str(array_to_write)+'\n'
elif dim == 1:
string = ' '.join(astr(array_to_write,prec=8))+'\n'
else:
raise ValueError('3_calculate_augmented_model_outputs.py: Only scalar and 1-dimensional model outputs are supported yet!')
return string
'Only scalar and 1D model outputs are supported!')
for ikey in range(nkeys):
for ipara in range(dims_all):
if ee_counter[ipara, ikey] > 0:
ee[ipara, ikey] /= ee_counter[ipara, ikey]
# -------------------------
# write final file
# -------------------------
# format:
# # model output #1: 'out1'
# # model output #2: 'out2'
# # ii para_name elemeffect(ii),ii=1:3,jj=1:1 counter(ii),ii=1:3,jj=1:1
# 1 'x_1' 0.53458196335158181 5
# 2 'x_2' 7.0822368906630215 5
# 3 'x_3' 3.5460086652980554 5
f = open(outfile, 'w')
for ikey in range(nkeys):
f.write('# model output #' + str(ikey + 1) + ': ' + keys[ikey] + '\n')
f.write('# ii para_name elemeffect(ii),ii=1:' + str(dims_all) +
',jj=1:' + str(nkeys) + ' counter(ii),ii=1:' + str(dims_all) +
',jj=1:' + str(nkeys) + ' \n')
for ipara in range(dims_all):
f.write(
str(ipara) + ' ' + para_name[ipara] + ' ' +
' '.join(astr(ee[ipara, :], prec=8)) + ' ' +
' '.join(astr(ee_counter[ipara, :])) + '\n')
f.close()
print("wrote: '" + outfile + "'")
figsize = mpl.rcParams['figure.figsize']
ifig = 0
# -------------------------------------------------------------------------
# Fig 1 - sorted Elementary Effects
#
if (multi_obj_approach == 'rectangle'):
keepit = []
cutoff_obj = np.ones(nobj) * -9999.
for iobj in range(nobj):
sort_idx = np.argsort(ee_masked[:, iobj])
print('')
print('OBJECTIVE #', iobj + 1)
print('sorted ee: ', astr(ee_masked[sort_idx, iobj], prec=4))
print('correspond to para: ', idx_para[sort_idx] + 1)
if (not (noplot)):
ifig += 1
iplot = 0
# print(' Plot - Fig ', ifig)
fig = plt.figure(ifig)
if (not (noplot)):
iplot += 1
ylab = r'$\mathrm{EE}$'
sub = fig.add_axes(
position(nrow, ncol, iplot, hspace=hspace, vspace=vspace))
if cutoff == '-1':
cbar.set_ticks(cticks)
cbar.set_ticklabels(cticknames)
cbar.ax.tick_params(direction='out', length=3, width=1, colors=fgcolor, labelsize=cbartsize, pad=-1.0,
bottom='off', top='on', labelbottom='off', labeltop='on')
else:
cticks = cvals[:-1] # remove last tick
cticks = cvals # dont remove last tick
cticknames = [ str2tex('$C_{'+str(i)+'}$') if i>0 else str2tex('no data') for i in cticks ]
cticknames = str2tex(cticknames, usetex=usetex)
cbar.set_ticks(cticks[:-1]+np.diff(cticks)/2.) # draw ticks in the middle
cbar.set_ticklabels(cticknames)
cbar.ax.tick_params(direction='out', length=3, width=1, colors=fgcolor, labelsize=cbartsize, pad=-1.0,
bottom=False, top=True, labelbottom=False, labeltop=True)
for iitick,itick in enumerate(cticks[:-1]):
tick_location = cticks
csub.text(1.0/(len(cvals)-1)*(iitick+0.5), 0.5, str2tex(astr(percent_in_cat[iitick],prec=1)+'%',usetex=usetex),
fontsize='xx-small', va='center', ha='center',rotation=0, transform=csub.transAxes)
# categorizer name
if categorizer_name == 'categorize_meteo_decision_tree_EEE':
clab = "EEE-based\n category"
elif categorizer_name == 'categorize_meteo_decision_tree_expert':
clab = "Expert-based\n category"
else:
clab = 'Not known\n category label'
clab = str2tex(clab, usetex=usetex)
cbar.ax.yaxis.set_label_text(clab, fontsize=cbartsize, va='center', ha='right', rotation=0)
cbar.ax.yaxis.set_label_coords(-0.03, 0.5, transform=csub.transAxes)
dy = -0.5
input_file=input_file,
time_step=time_step,
date_time_step=date_time_step,
period_time_step=period_time_step,
all_time_steps=all_time_steps,
grid_cells=grid_cells,
recipe=recipe)
if all_time_steps:
count = {
cc: np.shape(np.where(categories == cc)[0])[0]
for cc in np.unique(categories)
}
percentage = {
cc: np.shape(np.where(categories == cc)[0])[0] /
np.float(len(categories)) * 100.0
for cc in np.unique(categories)
}
print("count: ", count)
print("percentage: ", percentage)
print("")
for cc in np.unique(categories):
print(
str(count[cc]) + " & (" + astr(percentage[cc], prec=0) +
"\%) & ")
else:
print("")
print('category: ', categories, ' (numbers start with 1)')
# if informative(0) -> maskpara=False
# if noninformative(1) -> maskpara=True
mask_para = np.where((nc[:, 3].flatten()) == 1., True, False)
dims_all = np.shape(mask_para)[0]
idx_para = np.arange(dims_all)[
mask_para] # indexes of parameters which will be changed [0,npara-1]
dims = np.sum(mask_para)
# pick only non-masked bounds
lower_bound_mask = lower_bound[np.where(mask_para)]
upper_bound_mask = upper_bound[np.where(mask_para)]
para_dist_mask = para_dist[np.where(mask_para)]
para_name_mask = para_name[np.where(mask_para)]
fileID = astr(np.arange(1, nfiles + 1), zero=True)
# print('dims = '+astr(dims))
for kk in range(nfiles):
# print('Sampling #'+astr(kk+1)+' of '+astr(nfiles))
lower_bound_01 = np.zeros(dims)
upper_bound_01 = np.ones(dims)
[OptMatrix, OptOutVec] = morris_sampling(dims,
lower_bound_01,
upper_bound_01,
N=ntraj,
p=6,
r=ntraj,
keys = sobol_indexes_raven['paras']['si'].keys()
print("")
for ikey in keys:
print("")
print("---------------------")
print(ikey)
print("---------------------")
# scalar model output (such as NSE)
if (len(np.shape(
sobol_indexes_raven['paras']['si'][ikey])) == 1):
# parameter sensitivities
print(" si numeric (x1,x2,x3,...,w1,w2,w3,...) = ", [
astr(pp, prec=5) if ~np.isnan(pp) else "nan"
for pp in sobol_indexes_raven['paras']['si'][ikey]
])
print(" sti numeric (x1,x2,x3,...,w1,w2,w3,...) = ", [
astr(pp, prec=5) if ~np.isnan(pp) else "nan"
for pp in sobol_indexes_raven['paras']['sti'][ikey]
])
# process option sensitivities
print(" si numeric (a1,a2,a3,...,w1,w2,w3,...) = ", [
astr(pp, prec=5) if ~np.isnan(pp) else "nan" for pp in
sobol_indexes_raven['process_options']['si'][ikey]
])
print(" sti numeric (a1,a2,a3,...,w1,w2,w3,...) = ", [
astr(pp, prec=5) if ~np.isnan(pp) else "nan" for pp in
sobol_indexes_raven['process_options']['sti'][ikey]
jset)
sobol_indexes_disjoint_tmp[str(
jset
)] = sa_model_multiple_processes_DVM.sa_model_multiple_processes_DVM(
options_paras_disjoint,
para_ranges,
model_function_disjoint,
constants=None,
nsets=iset,
nsets_parasets=jset)
# process sensitivities
print(
" si numeric (A,B,C) = ",
astr(sobol_indexes_disjoint_tmp[str(jset)]['processes']
['si']['out'],
prec=5))
print(
" si analytic (A,B,C) = ",
astr(theo_si_sti_disjoint['wa_A-wb_B-wc_C_process'][0],
prec=5))
print(
" sti numeric (A,B,C) = ",
astr(sobol_indexes_disjoint_tmp[str(jset)]['processes']
['sti']['out'],
prec=5))
print(
" sti analytic (A,B,C) = ",
astr(theo_si_sti_disjoint['wa_A-wb_B-wc_C_process'][1],
prec=5))
print(
lsub.set_xticks([])
lsub.set_yticks([])
lsub.set_axis_off()
if (outtype == 'pdf'):
pdf_pages.savefig(fig)
plt.close(fig)
elif (outtype == 'png'):
pngfile = pngbase + "{0:04d}".format(ifig) + ".png"
fig.savefig(pngfile,
transparent=transparent,
bbox_inches=bbox_inches,
pad_inches=pad_inches)
plt.close(fig)
# -------------------------------------------------------------------------
# Finished
#
if (outtype == 'pdf'):
pdf_pages.close()
elif (outtype == 'png'):
pass
else:
plt.show()
t2 = time.time()
strin = '[m]: ' + astr(
(t2 - t1) / 60., 1) if (t2 - t1) > 60. else '[s]: ' + astr(t2 - t1, 0)
print('Time ', strin)
horizontalalignment='center',
verticalalignment='bottom',
transform=sub.transAxes)
if (outtype == 'pdf'):
pdf_pages.savefig(fig)
plt.close(fig)
elif (outtype == 'png'):
pngfile = pngbase + "{0:04d}".format(ifig) + ".png"
fig.savefig(pngfile,
transparent=transparent,
bbox_inches=bbox_inches,
pad_inches=pad_inches)
plt.close(fig)
# --------------------------------------
# Finish
# --------------------------------------
if (outtype == 'pdf'):
pdf_pages.close()
elif (outtype == 'png'):
pass
else:
plt.show()
t2 = time.time()
str = ' Time plot [m]: ' + astr(
(t2 - t1) /
60., 1) if (t2 - t1) > 60. else ' Time plot [s]: ' + astr(t2 - t1, 0)
print(str)
plt.setp(sub, xlabel=xlab) # axis labels
plt.setp(sub, ylabel=ylab)
sub.grid(False)
sub.text(0.0,
1.14,
str2tex("Original: "),
transform=sub.transAxes,
color=mcol2,
rotation=0,
fontsize='small',
horizontalalignment='left',
verticalalignment='center')
sub.text(0.17,
1.14,
str2tex("NSE = " + astr(nse_original[ibasin], prec=2),
usetex=usetex),
transform=sub.transAxes,
color=mcol2,
rotation=0,
fontsize='small',
horizontalalignment='left',
verticalalignment='center')
sub.text(0.34,
1.14,
str2tex("KGE = " + astr(kge_original[ibasin], prec=2),
usetex=usetex),
transform=sub.transAxes,
color=mcol2,
rotation=0,
fontsize='small',
for iset in nsets:
print('')
print('SA of disjoint-parameter setup: nsets = ',iset)
basin_prop={'id':'xxx'}
sobol_indexes_disjoint[str(iset)] = sa_model_multiple_processes.sa_model_multiple_processes(options_paras_disjoint,
para_ranges,
model_function_disjoint,
basin_prop,
constants=None,
nsets=iset)
# parameter sensitivities
print(" si numeric (x1,x2,x3,x4,x5,x6,x7,w1,..,w4) = ",astr(sobol_indexes_disjoint[str(iset)]['paras']['si']['out'],prec=5))
print(" si analytic (x1,x2,x3,x4,x5,x6,x7,w1,..,w4) = ",astr(theo_si_sti_disjoint['wa_A-wb_B-wc_C_para'][0],prec=5))
print(" sti numeric (x1,x2,x3,x4,x5,x6,x7,w1,..,w4) = ",astr(sobol_indexes_disjoint[str(iset)]['paras']['sti']['out'],prec=5))
print(" sti analytic (x1,x2,x3,x4,x5,x6,x7,w1,..,w4) = ",astr(theo_si_sti_disjoint['wa_A-wb_B-wc_C_para'][1],prec=5))
print(" mae error (x1,x2,x3,x4,x5,x6,x7,w1,..,w4) = ",np.mean(np.array(
list(np.abs(sobol_indexes_disjoint[str(iset)]['paras']['si']['out']-theo_si_sti_disjoint['wa_A-wb_B-wc_C_para'][0]))+
list(np.abs(sobol_indexes_disjoint[str(iset)]['paras']['sti']['out']-theo_si_sti_disjoint['wa_A-wb_B-wc_C_para'][1])))))
# process option sensitivities
print(" si numeric (a1,a2,b1,b2,b3,c1,c2,w1,..,w4) = ",astr(sobol_indexes_disjoint[str(iset)]['process_options']['si']['out'],prec=5))
print(" si analytic (a1,a2,b1,b2,b3,c1,c2,w1,..,w4) = ",astr(theo_si_sti_disjoint['wa_A-wb_B-wc_C_option'][0],prec=5))
print(" sti numeric (a1,a2,b1,b2,b3,c1,c2,w1,..,w4) = ",astr(sobol_indexes_disjoint[str(iset)]['process_options']['sti']['out'],prec=5))
print(" sti analytic (a1,a2,b1,b2,b3,c1,c2,w1,..,w4) = ",astr(theo_si_sti_disjoint['wa_A-wb_B-wc_C_option'][1],prec=5))
print(" mae error (a1,a2,b1,b2,b3,c1,c2,w1,..,w4) = ",np.mean(np.array(
list(np.abs(sobol_indexes_disjoint[str(iset)]['process_options']['si']['out']-theo_si_sti_disjoint['wa_A-wb_B-wc_C_option'][0]))+
list(np.abs(sobol_indexes_disjoint[str(iset)]['process_options']['sti']['out']-theo_si_sti_disjoint['wa_A-wb_B-wc_C_option'][1])))))
# I unsigned int integer 4
# l long integer 4
# L unsigned long integer 4
# f float float 4
# d double float 8
# s char[] string
from_day = 0
to_day = 0
#
for iyear in range(startyear, endyear + 1):
leap = (((iyear % 4) == 0) & ((iyear % 100) != 0)) | ((iyear % 400) == 0)
days_year = 365 + leap
bindata = open(indir + str(iyear) + '.bin', "rb").read()
values = np.array(
struct.unpack(
astr(ncols * nrows * days_year) + 'f',
bindata[0:4 * ncols * nrows * days_year]))
# create daily fields and roll axis, because for netcdf writer first dimension has to be time
var = np.rollaxis(
np.reshape(values, (nrows, ncols, days_year), order='Fortran'), 2, 0)
from_day = to_day
to_day = to_day + days_year
# write data to nc file
writenetcdf(fhandle, vhandle, var=var, time=np.arange(from_day, to_day))
# write time steps to nc
times = np.arange(from_day, to_day)
writenetcdf(fhandle,
thand,
time=np.arange(times[0], times[-1] + 1),
var=times)
# close netcdf
elif ( len(np.shape(model_output[ikey])) == 2):
# 1D model output
ee[ipara_changed,ikey] += np.mean(np.abs(model_output[ikey][iset,:]-model_output[ikey][iset+1,:]) / np.abs(parasets[iset,ipara_changed] - parasets[iset+1,ipara_changed]))
else:
raise ValueError('Only scalar and 1D model outputs are supported!')
for ikey in range(nkeys):
for ipara in range(dims_all):
if ee_counter[ipara,ikey] > 0:
ee[ipara,ikey] /= ee_counter[ipara,ikey]
# -------------------------
# write final file
# -------------------------
# format:
# # model output #1: 'out1'
# # model output #2: 'out2'
# # ii para_name elemeffect(ii),ii=1:3,jj=1:1 counter(ii),ii=1:3,jj=1:1
# 1 'x_1' 0.53458196335158181 5
# 2 'x_2' 7.0822368906630215 5
# 3 'x_3' 3.5460086652980554 5
f = open(outfile, 'w')
for ikey in range(nkeys):
f.write('# model output #'+str(ikey+1)+': '+keys[ikey]+'\n')
f.write('# ii para_name elemeffect(ii),ii=1:'+str(dims_all)+',jj=1:'+str(nkeys)+' counter(ii),ii=1:'+str(dims_all)+',jj=1:'+str(nkeys)+' \n')
for ipara in range(dims_all):
f.write(str(ipara)+' '+para_name[ipara]+' '+' '.join(astr(ee[ipara,:],prec=8))+' '+' '.join(astr(ee_counter[ipara,:]))+'\n')
f.close()
print("wrote: '"+outfile+"'")