def loss_function(args):
tau_mean, const_mean, tage_mean, fburst_mean, tburst_mean, logzsol_mean, gas_logz_mean, gas_logu_mean,\
tau_sig, const_sig, tage_sig, fburst_sig, tburst_sig, logzsol_sig, gas_logz_sig, gas_logu_sig = args
set_size = 3000
tau_arr = [
float(
priors.ClippedNormal(mean=abs(tau_mean),
sigma=abs(tau_sig),
mini=1.0,
maxi=8.0).sample()) for _ in range(set_size)
]
const_arr = [
float(
priors.ClippedNormal(mean=abs(const_mean),
sigma=abs(const_sig),
mini=0.0,
maxi=0.5).sample()) for _ in range(set_size)
]
tage_arr = [
float(
priors.ClippedNormal(mean=abs(tage_mean),
sigma=abs(tage_sig),
mini=1.0,
maxi=11.0).sample()) for _ in range(set_size)
]
fburst_arr = [
float(
priors.ClippedNormal(mean=abs(fburst_mean),
sigma=abs(fburst_sig),
mini=0.0,
maxi=0.8).sample()) for _ in range(set_size)
]
tburst_arr = [
float(
priors.ClippedNormal(mean=abs(tburst_mean),
sigma=abs(tburst_sig),
mini=0.0,
maxi=abs(tage_mean)).sample())
for _ in range(set_size)
]
logzsol_arr = [
float(
priors.ClippedNormal(mean=-1 * abs(logzsol_mean),
sigma=abs(logzsol_sig),
mini=-1.5,
maxi=0.0).sample()) for _ in range(set_size)
]
gas_logz_arr = [
float(
priors.ClippedNormal(mean=-1 * abs(gas_logz_mean),
sigma=abs(gas_logz_sig),
mini=-1.5,
maxi=0.0).sample()) for _ in range(set_size)
]
gas_logu_arr = [
float(
priors.ClippedNormal(mean=-1 * abs(gas_logu_mean),
sigma=abs(gas_logu_sig),
mini=-4.0,
maxi=-1.0).sample()) for _ in range(set_size)
]
# Fix the fburst + const > 1 issue
for ii in np.arange(len(const_arr)):
if const_arr[ii] + fburst_arr[ii] >= 0.95:
f_over = (const_arr[ii] + fburst_arr[ii]) - 0.95
if fburst_arr[ii] >= (f_over + 0.01):
fburst_arr[ii] = fburst_arr[ii] - (f_over + 0.01)
else:
const_arr[ii] = const_arr[ii] - (f_over + 0.01)
# Fixed the rest
dust1_arr = np.full(set_size, 0.1)
dust2_arr = np.full(set_size, 0.0)
sf_trunc_arr = np.full(set_size, 0.0)
# List of model parameters
dwarf_sample_parameters = [{
'dust1': dust1_arr[ii],
'dust2': dust2_arr[ii],
'logzsol': logzsol_arr[ii],
'gas_logz': gas_logz_arr[ii],
'gas_logu': gas_logu_arr[ii],
'const': const_arr[ii],
'tau': tau_arr[ii],
'tage': tage_arr[ii],
'sf_trunc': sf_trunc_arr[ii],
'fburst': fburst_arr[ii],
'tburst': tburst_arr[ii]
} for ii in np.arange(set_size)]
# Double check
for ii, model in enumerate(dwarf_sample_parameters):
if model['fburst'] + model['const'] >= 0.99:
print(ii, model['fburst'], model['const'])
# Initialize the spop model
spop_tau = setup_fsps_spop(zcontinuous=1,
imf_type=2,
sfh=1,
dust_type=0,
dust_index=-1.3,
dust1_index=-1.0)
# Get the SDSS filters
sdss_bands = fsps.find_filter('SDSS')
dwarf_sample_gaussian = generate_dwarf_population(spop_tau,
dwarf_sample_parameters,
filters=sdss_bands,
n_jobs=6)
# Measure colors and emission line EWs
# - SDSS_EMLINES is a pre-defined dict of emission lines center wavelength and the
# wavelength window for measuring EW.
# - You can save the results in a numpy array
dwarf_sample_table = measure_color_ew(dwarf_sample_gaussian,
em_list=SDSS_EMLINES,
output=None)
bin_size = 200
ur_size = np.linspace(0.0, 2.5, bin_size)
ug_size = np.linspace(0.0, 2.0, bin_size)
gr_size = np.linspace(-0.1, 0.8, bin_size)
gi_size = np.linspace(-0.2, 1.3, bin_size)
ha_size = np.linspace(0.0, 3.0, bin_size)
hb_size = np.linspace(-0.5, 2.5, bin_size)
oiii_size = np.linspace(-1.0, 3.0, bin_size)
obs_ur = data_to_distribution(
np.asarray(sdss_use['M_u'] - sdss_use['M_r']), ur_size)
obs_ug = data_to_distribution(
np.asarray(sdss_use['M_u'] - sdss_use['M_g']), ug_size)
obs_gr = data_to_distribution(
np.asarray(sdss_use['M_g'] - sdss_use['M_r']), gr_size)
obs_gi = data_to_distribution(
np.asarray(sdss_use['M_g'] - sdss_use['M_i']), gi_size)
obs_ha = data_to_distribution(np.log10(abs(sdss_use['H_ALPHA_EQW'])),
ha_size)
obs_hb = data_to_distribution(np.log10(abs(sdss_use['H_BETA_EQW'])),
hb_size)
obs_oiii = data_to_distribution(np.log10(abs(sdss_use['OIII_5007_EQW'])),
oiii_size)
model_ur = data_to_distribution(dwarf_sample_table['ur_color'], ur_size)
model_ug = data_to_distribution(dwarf_sample_table['ug_color'], ug_size)
model_gr = data_to_distribution(dwarf_sample_table['gr_color'], gr_size)
model_gi = data_to_distribution(dwarf_sample_table['gi_color'], gi_size)
model_ha = data_to_distribution(
np.log10(abs(dwarf_sample_table['ew_halpha'])), ha_size)
model_hb = data_to_distribution(
np.log10(abs(dwarf_sample_table['ew_hbeta'])), hb_size)
model_oiii = data_to_distribution(
np.log10(abs(dwarf_sample_table['ew_oiii_5007'])), oiii_size)
obs_stack = np.transpose(
np.vstack([obs_ur, obs_ug, obs_gr, obs_gi, obs_ha, obs_hb, obs_oiii]))
model_stack = np.transpose(
np.vstack([
model_ur, model_ug, model_gr, model_gi, model_ha, model_hb,
model_oiii
]))
total_loss = entropy(obs=obs_stack, model=model_stack)
return total_loss
def loss_function(args):
tau_mean, const_mean, tage_mean, fburst_mean, tburst_mean, logzsol_mean, gas_logz_mean, gas_logu_mean = args
set_size = 3000
tau_arr = [
float(
priors.ClippedNormal(mean=tau_mean,
sigma=tau_sig,
mini=1.0,
maxi=8.0).sample()) for _ in range(set_size)
]
const_arr = [
float(
priors.ClippedNormal(mean=const_mean,
sigma=const_sig,
mini=0.0,
maxi=0.5).sample()) for _ in range(set_size)
]
tage_arr = [
float(
priors.ClippedNormal(mean=tage_mean,
sigma=tage_sig,
mini=1.0,
maxi=11.0).sample()) for _ in range(set_size)
]
fburst_arr = [
float(
priors.ClippedNormal(mean=fburst_mean,
sigma=fbust_sig,
mini=0.0,
maxi=0.8).sample()) for _ in range(set_size)
]
tburst_arr = [
float(
priors.ClippedNormal(mean=tburst_mean,
sigma=tburst_sig,
mini=0.0,
maxi=8.0).sample()) for _ in range(set_size)
]
logzsol_arr = [
float(
priors.ClippedNormal(mean=logzsol_mean,
sigma=logzsol_sig,
mini=-1.5,
maxi=0.0).sample()) for _ in range(set_size)
]
gas_logz_arr = [
float(
priors.ClippedNormal(mean=gas_logz_mean,
sigma=gas_logz_sig,
mini=-1.5,
maxi=0.0).sample()) for _ in range(set_size)
]
gas_logu_arr = [
float(
priors.ClippedNormal(mean=gas_logu_mean,
sigma=gas_logu_sig,
mini=-4.0,
maxi=-1.0).sample()) for _ in range(set_size)
]
# Fix the fburst + const > 1 issue
for ii in np.arange(len(const_arr)):
if const_arr[ii] + fburst_arr[ii] >= 0.95:
f_over = (const_arr[ii] + fburst_arr[ii]) - 0.95
if fburst_arr[ii] >= (f_over + 0.01):
fburst_arr[ii] = fburst_arr[ii] - (f_over + 0.01)
else:
const_arr[ii] = const_arr[ii] - (f_over + 0.01)
# Fixed the rest
dust1_arr = np.full(set_size, 0.1)
dust2_arr = np.full(set_size, 0.0)
sf_trunc_arr = np.full(set_size, 0.0)
# List of model parameters
dwarf_sample_parameters = [{
'dust1': dust1_arr[ii],
'dust2': dust2_arr[ii],
'logzsol': logzsol_arr[ii],
'gas_logz': gas_logz_arr[ii],
'gas_logu': gas_logu_arr[ii],
'const': const_arr[ii],
'tau': tau_arr[ii],
'tage': tage_arr[ii],
'sf_trunc': sf_trunc_arr[ii],
'fburst': fburst_arr[ii],
'tburst': tburst_arr[ii]
} for ii in np.arange(set_size)]
# Double check
for ii, model in enumerate(dwarf_sample_parameters):
if model['fburst'] + model['const'] >= 0.99:
print(ii, model['fburst'], model['const'])
# Initialize the spop model
spop_tau = setup_fsps_spop(zcontinuous=1,
imf_type=2,
sfh=1,
dust_type=0,
dust_index=-1.3,
dust1_index=-1.0)
# Get the SDSS filters
sdss_bands = fsps.find_filter('SDSS')
dwarf_sample_gaussian = generate_dwarf_population(spop_tau,
dwarf_sample_parameters,
filters=sdss_bands,
n_jobs=4)
# Measure colors and emission line EWs
# - SDSS_EMLINES is a pre-defined dict of emission lines center wavelength and the
# wavelength window for measuring EW.
# - You can save the results in a numpy array
dwarf_sample_table = measure_color_ew(dwarf_sample_gaussian,
em_list=SDSS_EMLINES,
output=None)
bin_size = 200
ur_loss = loss(dwarf_sample_table['ur_color'],
np.asarray(sdss_use['M_u'] - sdss_use['M_r']),
np.linspace(0, 2.5, bin_size))
ug_loss = loss(dwarf_sample_table['ug_color'],
np.asarray(sdss_use['M_u'] - sdss_use['M_g']),
np.linspace(0, 1.75, bin_size))
gr_loss = loss(dwarf_sample_table['gr_color'],
np.asarray(sdss_use['M_g'] - sdss_use['M_r']),
np.linspace(-0.1, 0.8, bin_size))
gi_loss = loss(dwarf_sample_table['gi_color'],
np.asarray(sdss_use['M_g'] - sdss_use['M_i']),
np.linspace(-0.2, 1.2, bin_size))
OIII_loss = loss(np.log10(dwarf_sample_table['ew_oiii_5007']),
np.log10(-1.0 * (sdss_use['OIII_5007_EQW'])),
np.linspace(-1, 3, bin_size))
Ha_loss = loss(np.log10(np.log10(dwarf_sample_table['ew_halpha'])),
np.log10(-1.0 * sdss_use['H_ALPHA_EQW']),
np.linspace(0, 3, bin_size))
Hb_loss = loss(np.log10(dwarf_sample_table['ew_hbeta']),
np.log10(-1.0 * sdss_use['H_BETA_EQW']),
np.linspace(-0.5, 2.5, bin_size))
total_loss = (ur_loss + ug_loss + gr_loss + gi_loss + OIII_loss + Ha_loss +
Hb_loss) / 7
part_loss = [
ur_loss, ug_loss, gr_loss, gi_loss, OIII_loss, Ha_loss, Hb_loss
]
return total_loss