elif opts.doLuminosity:
names = opts.name.split(",")
filenames = []
legend_names = []
for name in names:
for ii, model in enumerate(models):
filename = '%s/%s/%s_Lbol.dat' % (outputDir, model, name)
if not os.path.isfile(filename):
continue
filenames.append(filename)
legend_names.append(models_ref[ii])
break
Lbols, names = lightcurve_utils.read_files_lbol(filenames)
if opts.doModels:
model_data = {}
modelfiles = opts.modelfile.split(",")
for modelfile in modelfiles:
modelfile = os.path.join(opts.plotDir, modelfile)
modelfileSplit = modelfile.split("/")
model_out = np.loadtxt(modelfile)
errorbudget = float(modelfileSplit[-2])
modelType = modelfileSplit[-4]
model = modelfileSplit[-6].split("_")[0]
if model == "BNS" and modelType == "ejecta":
def calc_svd_lbol(tini,tmax,dt, n_coeff = 100, model = "BaKa2016"):
print("Calculating SVD model of bolometric luminosity...")
if model == "BaKa2016":
fileDir = "../output/barnes_kilonova_spectra"
elif model == "Ka2017":
fileDir = "../output/kasen_kilonova_grid"
elif model == "RoFe2017":
fileDir = "../output/macronovae-rosswog_wind"
filenames = glob.glob('%s/*_Lbol.dat'%fileDir)
lbols, names = lightcurve_utils.read_files_lbol(filenames)
lbolkeys = lbols.keys()
tt = np.arange(tini,tmax+dt,dt)
for key in lbolkeys:
keySplit = key.split("_")
if keySplit[0] == "rpft":
mej0 = float("0." + keySplit[1].replace("m",""))
vej0 = float("0." + keySplit[2].replace("v",""))
lbols[key]["mej"] = mej0
lbols[key]["vej"] = vej0
elif keySplit[0] == "knova":
mej0 = float(keySplit[3].replace("m",""))
vej0 = float(keySplit[4].replace("vk",""))
if len(keySplit) == 6:
Xlan0 = 10**float(keySplit[5].replace("Xlan1e",""))
elif len(keySplit) == 7:
if "Xlan1e" in keySplit[6]:
Xlan0 = 10**float(keySplit[6].replace("Xlan1e",""))
elif "Xlan1e" in keySplit[5]:
Xlan0 = 10**float(keySplit[5].replace("Xlan1e",""))
#if (mej0 == 0.05) and (vej0 == 0.2) and (Xlan0 == 1e-3):
# del lbols[key]
# continue
lbols[key]["mej"] = mej0
lbols[key]["vej"] = vej0
lbols[key]["Xlan"] = Xlan0
elif keySplit[0] == "SED":
lbols[key]["mej"], lbols[key]["vej"], lbols[key]["Ye"] = lightcurve_utils.get_macronovae_rosswog(key)
ii = np.where(np.isfinite(lbols[key]["Lbol"]))[0]
f = interp.interp1d(lbols[key]["tt"][ii], np.log10(lbols[key]["Lbol"][ii]), fill_value='extrapolate')
lbolinterp = 10**f(tt)
lbols[key]["Lbol"]= np.log10(lbolinterp)
lbolkeys = lbols.keys()
lbol_array = []
param_array = []
for key in lbolkeys:
lbol_array.append(lbols[key]["Lbol"])
if model == "BaKa2016":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["vej"]])
elif model == "Ka2017":
param_array.append([np.log10(lbols[key]["mej"]),np.log10(lbols[key]["vej"]),np.log10(lbols[key]["Xlan"])])
elif model == "RoFe2017":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["vej"],lbols[key]["Ye"]])
param_array_postprocess = np.array(param_array)
param_mins, param_maxs = np.min(param_array_postprocess,axis=0),np.max(param_array_postprocess,axis=0)
for i in range(len(param_mins)):
param_array_postprocess[:,i] = (param_array_postprocess[:,i]-param_mins[i])/(param_maxs[i]-param_mins[i])
lbol_array_postprocess = np.array(lbol_array)
mins,maxs = np.min(lbol_array_postprocess,axis=0),np.max(lbol_array_postprocess,axis=0)
for i in range(len(mins)):
lbol_array_postprocess[:,i] = (lbol_array_postprocess[:,i]-mins[i])/(maxs[i]-mins[i])
lbol_array_postprocess[np.isnan(lbol_array_postprocess)]=0.0
UA, sA, VA = np.linalg.svd(lbol_array_postprocess, full_matrices=True)
VA = VA.T
n, n = UA.shape
m, m = VA.shape
cAmat = np.zeros((n_coeff,n))
cAvar = np.zeros((n_coeff,n))
for i in range(n):
cAmat[:,i] = np.dot(lbol_array_postprocess[i,:],VA[:,:n_coeff])
ErrorLevel = 2.0
errors = ErrorLevel*lbol_array_postprocess[i,:]
cAvar[:,i] = np.diag(np.dot(VA[:,:n_coeff].T,np.dot(np.diag(np.power(errors,2.)),VA[:,:n_coeff])))
cAstd = np.sqrt(cAvar)
nsvds, nparams = param_array_postprocess.shape
kernel = 1.0 * RationalQuadratic(length_scale=1.0, alpha=0.1)
gps = []
for i in range(n_coeff):
gp = GaussianProcessRegressor(kernel=kernel,n_restarts_optimizer=0)
gp.fit(param_array_postprocess, cAmat[i,:])
gps.append(gp)
svd_model = {}
svd_model["n_coeff"] = n_coeff
svd_model["param_array"] = param_array
svd_model["cAmat"] = cAmat
svd_model["cAstd"] = cAstd
svd_model["VA"] = VA
svd_model["param_mins"] = param_mins
svd_model["param_maxs"] = param_maxs
svd_model["mins"] = mins
svd_model["maxs"] = maxs
svd_model["gps"] = gps
svd_model["tt"] = tt
print("Finished calculating SVD model of bolometric luminosity...")
return svd_model
def calc_svd_lbol(tini, tmax, dt, n_coeff=100, model="BaKa2016"):
print("Calculating SVD model of bolometric luminosity...")
if model == "BaKa2016":
fileDir = "../output/barnes_kilonova_spectra"
elif model == "Ka2017":
fileDir = "../output/kasen_kilonova_grid"
elif model == "RoFe2017":
fileDir = "../output/macronovae-rosswog_wind"
elif model == "Bu2019":
fileDir = "../output/bulla_1D"
fileDir = "../output/bulla_1D_phi0"
fileDir = "../output/bulla_1D_phi90"
elif model == "Bu2019inc":
fileDir = "../output/bulla_2D"
elif model == "Bu2019lf":
fileDir = "../output/bulla_2Component_lfree"
elif model == "Bu2019lr":
fileDir = "../output/bulla_2Component_lrich"
elif model == "Bu2019lm":
fileDir = "../output/bulla_2Component_lmid"
elif model == "Bu2019lw":
fileDir = "../output/bulla_2Component_lmid_0p005"
elif model == "Bu2019bc":
fileDir = "../output/bulla_blue_cone"
elif model == "Bu2019re":
fileDir = "../output/bulla_red_ellipse"
elif model == "Bu2019op":
fileDir = "../output/bulla_opacity"
elif model == "Bu2019ops":
fileDir = "../output/bulla_opacity_slim"
elif model == "Bu2019rp":
fileDir = "../output/bulla_reprocess"
elif model == "Bu2019rps":
fileDir = "../output/bulla_reprocess_slim"
filenames = glob.glob('%s/*_Lbol.dat' % fileDir)
lbols, names = lightcurve_utils.read_files_lbol(filenames)
lbolkeys = lbols.keys()
tt = np.arange(tini, tmax + dt, dt)
for key in lbolkeys:
keySplit = key.split("_")
if keySplit[0] == "rpft":
mej0 = float("0." + keySplit[1].replace("m", ""))
vej0 = float("0." + keySplit[2].replace("v", ""))
lbols[key]["mej"] = mej0
lbols[key]["vej"] = vej0
elif keySplit[0] == "knova":
mej0 = float(keySplit[3].replace("m", ""))
vej0 = float(keySplit[4].replace("vk", ""))
if len(keySplit) == 6:
Xlan0 = 10**float(keySplit[5].replace("Xlan1e", ""))
elif len(keySplit) == 7:
if "Xlan1e" in keySplit[6]:
Xlan0 = 10**float(keySplit[6].replace("Xlan1e", ""))
elif "Xlan1e" in keySplit[5]:
Xlan0 = 10**float(keySplit[5].replace("Xlan1e", ""))
#if (mej0 == 0.05) and (vej0 == 0.2) and (Xlan0 == 1e-3):
# del lbols[key]
# continue
lbols[key]["mej"] = mej0
lbols[key]["vej"] = vej0
lbols[key]["Xlan"] = Xlan0
elif keySplit[0] == "SED":
lbols[key]["mej"], lbols[key]["vej"], lbols[key][
"Ye"] = lightcurve_utils.get_macronovae_rosswog(key)
elif "gamma" in key:
kappaLF = float(keySplit[2].replace("kappaLF", ""))
gammaLF = float(keySplit[3].replace("gammaLF", ""))
kappaLR = float(keySplit[4].replace("kappaLR", ""))
gammaLR = float(keySplit[5].replace("gammaLR", ""))
theta = float(keySplit[6])
lbols[key]["kappaLF"] = kappaLF
lbols[key]["gammaLF"] = gammaLF
lbols[key]["kappaLR"] = kappaLR
lbols[key]["gammaLR"] = gammaLR
lbols[key]["theta"] = theta
elif "nsns" in key:
mejdyn = float(keySplit[2].replace("mejdyn", ""))
mejwind = float(keySplit[3].replace("mejwind", ""))
phi0 = float(keySplit[4].replace("phi", ""))
theta = float(keySplit[5])
lbols[key]["mej_dyn"] = mejdyn
lbols[key]["mej_wind"] = mejwind
lbols[key]["phi"] = phi0
lbols[key]["theta"] = theta
elif "mejdyn" in key:
mejdyn = float(keySplit[1].replace("mejdyn", ""))
mejwind = float(keySplit[2].replace("mejwind", ""))
phi0 = float(keySplit[4].replace("phi", ""))
theta = float(keySplit[5])
lbols[key]["mej_dyn"] = mejdyn
lbols[key]["mej_wind"] = mejwind
lbols[key]["phi"] = phi0
lbols[key]["theta"] = theta
elif "bluecone" in key:
mej = float(keySplit[2].replace("mej", ""))
phi0 = float(keySplit[3].replace("th", ""))
theta = float(keySplit[4])
lbols[key]["mej"] = mej
lbols[key]["phi"] = 90 - phi0
lbols[key]["theta"] = theta
elif "redellips" in key:
mej = float(keySplit[2].replace("mej", ""))
a0 = float(keySplit[3].replace("a", ""))
theta = float(keySplit[4])
lbols[key]["mej"] = mej
lbols[key]["a"] = a0
lbols[key]["theta"] = theta
elif keySplit[0] == "nph1.0e+06":
mej0 = float(keySplit[1].replace("mej", ""))
phi0 = float(keySplit[2].replace("phi", ""))
theta = float(keySplit[3])
lbols[key]["mej"] = mej0
lbols[key]["phi"] = phi0
#lbols[key]["T"] = T0
lbols[key]["theta"] = theta
elif keySplit[0] == "kasenReprocess":
mej1 = float(keySplit[2].replace("mejcone", ""))
mej2 = float(keySplit[3].replace("mejell", ""))
phi = float(keySplit[4].replace("th", ""))
a = float(keySplit[5].replace("a", ""))
theta = float(keySplit[6])
lbols[key]["mej_1"] = mej1
lbols[key]["mej_2"] = mej2
lbols[key]["phi"] = phi
lbols[key]["a"] = a
lbols[key]["theta"] = theta
ii = np.where(np.isfinite(lbols[key]["Lbol"]))[0]
f = interp.interp1d(lbols[key]["tt"][ii],
np.log10(lbols[key]["Lbol"][ii]),
fill_value='extrapolate')
lbolinterp = 10**f(tt)
lbols[key]["Lbol"] = np.log10(lbolinterp)
lbolkeys = lbols.keys()
lbol_array = []
param_array = []
for key in lbolkeys:
lbol_array.append(lbols[key]["Lbol"])
if model == "BaKa2016":
param_array.append(
[np.log10(lbols[key]["mej"]), lbols[key]["vej"]])
elif model == "Ka2017":
param_array.append([
np.log10(lbols[key]["mej"]),
np.log10(lbols[key]["vej"]),
np.log10(lbols[key]["Xlan"])
])
elif model == "RoFe2017":
param_array.append([
np.log10(lbols[key]["mej"]), lbols[key]["vej"],
lbols[key]["Ye"]
])
elif model == "Bu2019":
param_array.append(
[np.log10(lbols[key]["mej"]),
np.log10(lbols[key]["T"])])
elif model == "Bu2019inc":
param_array.append([
np.log10(lbols[key]["mej"]), lbols[key]["phi"],
lbols[key]["theta"]
])
elif model in ["Bu2019lf", "Bu2019lr", "Bu2019lm"]:
param_array.append([
np.log10(lbols[key]["mej_dyn"]),
np.log10(lbols[key]["mej_wind"]), lbols[key]["phi"],
lbols[key]["theta"]
])
elif model in ["Bu2019lw"]:
param_array.append([
np.log10(lbols[key]["mej_wind"]), lbols[key]["phi"],
lbols[key]["theta"]
])
elif model == "Bu2019bc":
param_array.append([
np.log10(lbols[key]["mej"]), lbols[key]["phi"],
lbols[key]["theta"]
])
elif model == "Bu2019re":
param_array.append([
np.log10(lbols[key]["mej"]), lbols[key]["a"],
lbols[key]["theta"]
])
elif model == "Bu2019op":
param_array.append([
np.log10(lbols[key]["kappaLF"]), lbols[key]["gammaLF"],
np.log10(lbols[key]["kappaLR"]), lbols[key]["gammaLR"]
])
elif model == "Bu2019ops":
param_array.append([
np.log10(lbols[key]["kappaLF"]),
np.log10(lbols[key]["kappaLR"]), lbols[key]["gammaLR"]
])
elif model == "Bu2019rp":
param_array.append([
np.log10(lbols[key]["mej_1"]),
np.log10(lbols[key]["mej_2"]), lbols[key]["phi"],
lbols[key]["a"], lbols[key]["theta"]
])
elif model == "Bu2019rps":
param_array.append([
np.log10(lbols[key]["mej_1"]),
np.log10(lbols[key]["mej_2"]), lbols[key]["a"]
])
param_array_postprocess = np.array(param_array)
param_mins, param_maxs = np.min(param_array_postprocess,
axis=0), np.max(param_array_postprocess,
axis=0)
for i in range(len(param_mins)):
param_array_postprocess[:, i] = (param_array_postprocess[:, i] -
param_mins[i]) / (param_maxs[i] -
param_mins[i])
lbol_array_postprocess = np.array(lbol_array)
mins, maxs = np.min(lbol_array_postprocess,
axis=0), np.max(lbol_array_postprocess, axis=0)
for i in range(len(mins)):
lbol_array_postprocess[:, i] = (lbol_array_postprocess[:, i] -
mins[i]) / (maxs[i] - mins[i])
lbol_array_postprocess[np.isnan(lbol_array_postprocess)] = 0.0
UA, sA, VA = np.linalg.svd(lbol_array_postprocess, full_matrices=True)
VA = VA.T
n, n = UA.shape
m, m = VA.shape
cAmat = np.zeros((n_coeff, n))
cAvar = np.zeros((n_coeff, n))
for i in range(n):
cAmat[:, i] = np.dot(lbol_array_postprocess[i, :], VA[:, :n_coeff])
ErrorLevel = 2.0
errors = ErrorLevel * lbol_array_postprocess[i, :]
cAvar[:, i] = np.diag(
np.dot(VA[:, :n_coeff].T,
np.dot(np.diag(np.power(errors, 2.)), VA[:, :n_coeff])))
cAstd = np.sqrt(cAvar)
nsvds, nparams = param_array_postprocess.shape
kernel = 1.0 * RationalQuadratic(length_scale=1.0, alpha=0.1)
gps = []
for i in range(n_coeff):
gp = GaussianProcessRegressor(kernel=kernel, n_restarts_optimizer=0)
gp.fit(param_array_postprocess, cAmat[i, :])
gps.append(gp)
svd_model = {}
svd_model["n_coeff"] = n_coeff
svd_model["param_array"] = param_array
svd_model["cAmat"] = cAmat
svd_model["cAstd"] = cAstd
svd_model["VA"] = VA
svd_model["param_mins"] = param_mins
svd_model["param_maxs"] = param_maxs
svd_model["mins"] = mins
svd_model["maxs"] = maxs
svd_model["gps"] = gps
svd_model["tt"] = tt
print("Finished calculating SVD model of bolometric luminosity...")
return svd_model
def calc_svd_lbol(tini,tmax,dt, n_coeff = 100, model = "BaKa2016",
gptype="sklearn"):
print("Calculating SVD model of bolometric luminosity...")
if model == "BaKa2016":
fileDir = "../output/barnes_kilonova_spectra"
elif model == "Ka2017":
fileDir = "../output/kasen_kilonova_grid"
elif model == "RoFe2017":
fileDir = "../output/macronovae-rosswog_wind"
elif model == "Bu2019":
fileDir = "../output/bulla_1D"
fileDir = "../output/bulla_1D_phi0"
fileDir = "../output/bulla_1D_phi90"
elif model == "Bu2019inc":
fileDir = "../output/bulla_2D"
elif model == "Bu2019lf":
fileDir = "../output/bulla_2Component_lfree"
elif model == "Bu2019lr":
fileDir = "../output/bulla_2Component_lrich"
elif model == "Bu2019lm":
fileDir = "../output/bulla_2Component_lmid"
elif model == "Bu2019lw":
fileDir = "../output/bulla_2Component_lmid_0p005"
elif model == "Bu2019bc":
fileDir = "../output/bulla_blue_cone"
elif model == "Bu2019re":
fileDir = "../output/bulla_red_ellipse"
elif model == "Bu2019op":
fileDir = "../output/bulla_opacity"
elif model == "Bu2019ops":
fileDir = "../output/bulla_opacity_slim"
elif model == "Bu2019rp":
fileDir = "../output/bulla_reprocess"
elif model == "Bu2019rps":
fileDir = "../output/bulla_reprocess_slim"
elif model == "Bu2019rpd":
fileDir = "../output/bulla_reprocess_decimated"
elif model == "Bu2019nsbh":
fileDir = "../output/bulla_2Component_lnsbh"
elif model == "Wo2020dyn":
fileDir = "../output/bulla_rosswog_dynamical"
elif model == "Wo2020dw":
fileDir = "../output/bulla_rosswog_wind"
filenames = glob.glob('%s/*_Lbol.dat'%fileDir)
lbols, names = lightcurve_utils.read_files_lbol(filenames)
lbolkeys = lbols.keys()
tt = np.arange(tini,tmax+dt,dt)
for key in lbolkeys:
keySplit = key.split("_")
if keySplit[0] == "rpft":
mej0 = float("0." + keySplit[1].replace("m",""))
vej0 = float("0." + keySplit[2].replace("v",""))
lbols[key]["mej"] = mej0
lbols[key]["vej"] = vej0
elif keySplit[0] == "knova":
mej0 = float(keySplit[3].replace("m",""))
vej0 = float(keySplit[4].replace("vk",""))
if len(keySplit) == 6:
Xlan0 = 10**float(keySplit[5].replace("Xlan1e",""))
elif len(keySplit) == 7:
if "Xlan1e" in keySplit[6]:
Xlan0 = 10**float(keySplit[6].replace("Xlan1e",""))
elif "Xlan1e" in keySplit[5]:
Xlan0 = 10**float(keySplit[5].replace("Xlan1e",""))
#if (mej0 == 0.05) and (vej0 == 0.2) and (Xlan0 == 1e-3):
# del lbols[key]
# continue
lbols[key]["mej"] = mej0
lbols[key]["vej"] = vej0
lbols[key]["Xlan"] = Xlan0
elif keySplit[0] == "SED":
lbols[key]["mej"], lbols[key]["vej"], lbols[key]["Ye"] = lightcurve_utils.get_macronovae_rosswog(key)
elif "gamma" in key:
kappaLF = float(keySplit[2].replace("kappaLF",""))
gammaLF = float(keySplit[3].replace("gammaLF",""))
kappaLR = float(keySplit[4].replace("kappaLR",""))
gammaLR = float(keySplit[5].replace("gammaLR",""))
theta = float(keySplit[6])
lbols[key]["kappaLF"] = kappaLF
lbols[key]["gammaLF"] = gammaLF
lbols[key]["kappaLR"] = kappaLR
lbols[key]["gammaLR"] = gammaLR
lbols[key]["theta"] = theta
elif "nsns" in key:
mejdyn = float(keySplit[2].replace("mejdyn",""))
mejwind = float(keySplit[3].replace("mejwind",""))
phi0 = float(keySplit[4].replace("phi",""))
theta = float(keySplit[5])
lbols[key]["mej_dyn"] = mejdyn
lbols[key]["mej_wind"] = mejwind
lbols[key]["phi"] = phi0
lbols[key]["theta"] = theta
elif keySplit[0] == "nsbh":
mej_dyn = float(keySplit[2].replace("mejdyn",""))
mej_wind = float(keySplit[3].replace("mejwind",""))
phi = float(keySplit[4].replace("phi",""))
theta = float(keySplit[5])
lbols[key]["mej_dyn"] = mej_dyn
lbols[key]["mej_wind"] = mej_wind
#lbols[key]["phi"] = phi
lbols[key]["theta"] = theta
elif "mejdyn" in key:
mejdyn = float(keySplit[1].replace("mejdyn",""))
mejwind = float(keySplit[2].replace("mejwind",""))
phi0 = float(keySplit[4].replace("phi",""))
theta = float(keySplit[5])
lbols[key]["mej_dyn"] = mejdyn
lbols[key]["mej_wind"] = mejwind
lbols[key]["phi"] = phi0
lbols[key]["theta"] = theta
elif "bluecone" in key:
mej = float(keySplit[2].replace("mej",""))
phi0 = float(keySplit[3].replace("th",""))
theta = float(keySplit[4])
lbols[key]["mej"] = mej
lbols[key]["phi"] = 90-phi0
lbols[key]["theta"] = theta
elif "redellips" in key:
mej = float(keySplit[2].replace("mej",""))
a0 = float(keySplit[3].replace("a",""))
theta = float(keySplit[4])
lbols[key]["mej"] = mej
lbols[key]["a"] = a0
lbols[key]["theta"] = theta
elif keySplit[0] == "nph1.0e+06":
mej0 = float(keySplit[1].replace("mej",""))
phi0 = float(keySplit[2].replace("phi",""))
theta = float(keySplit[3])
lbols[key]["mej"] = mej0
lbols[key]["phi"] = phi0
#lbols[key]["T"] = T0
lbols[key]["theta"] = theta
elif keySplit[0] == "kasenReprocess":
mej1 = float(keySplit[2].replace("mejcone",""))
mej2 = float(keySplit[3].replace("mejell",""))
phi = float(keySplit[4].replace("th",""))
a = float(keySplit[5].replace("a",""))
theta = float(keySplit[6])
lbols[key]["mej_1"] = mej1
lbols[key]["mej_2"] = mej2
lbols[key]["phi"] = phi
lbols[key]["a"] = a
lbols[key]["theta"] = theta
elif keySplit[0] == "RGAdyn":
mej = float(keySplit[2].replace("mej",""))
a = float(keySplit[3].replace("a",""))
sd = float(keySplit[4].replace("sd",""))
theta = float(keySplit[5])
lbols[key]["mej"] = mej
lbols[key]["a"] = a
lbols[key]["sd"] = sd
lbols[key]["theta"] = theta
elif keySplit[0] == "WoWind":
mej = float(keySplit[2].replace("mej",""))
rwind = float(keySplit[3].replace("v",""))
theta = float(keySplit[4])
lbols[key]["mej"] = mej
lbols[key]["rwind"] = rwind
lbols[key]["theta"] = theta
ii = np.where(np.isfinite(lbols[key]["Lbol"]))[0]
f = interp.interp1d(lbols[key]["tt"][ii], np.log10(lbols[key]["Lbol"][ii]), fill_value='extrapolate')
lbolinterp = 10**f(tt)
lbols[key]["Lbol"]= np.log10(lbolinterp)
lbolkeys = lbols.keys()
lbol_array = []
param_array = []
for key in lbolkeys:
lbol_array.append(lbols[key]["Lbol"])
if model == "BaKa2016":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["vej"]])
elif model == "Ka2017":
param_array.append([np.log10(lbols[key]["mej"]),np.log10(lbols[key]["vej"]),np.log10(lbols[key]["Xlan"])])
elif model == "RoFe2017":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["vej"],lbols[key]["Ye"]])
elif model == "Bu2019":
param_array.append([np.log10(lbols[key]["mej"]),np.log10(lbols[key]["T"])])
elif model == "Bu2019inc":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["phi"],lbols[key]["theta"]])
elif model in ["Bu2019lf","Bu2019lr","Bu2019lm"]:
param_array.append([np.log10(lbols[key]["mej_dyn"]),np.log10(lbols[key]["mej_wind"]),lbols[key]["phi"],lbols[key]["theta"]])
elif model in ["Bu2019nsbh"]:
param_array.append([np.log10(lbols[key]["mej_dyn"]),np.log10(lbols[key]["mej_wind"]),lbols[key]["theta"]])
elif model in ["Bu2019lw"]:
param_array.append([np.log10(lbols[key]["mej_wind"]),lbols[key]["phi"],lbols[key]["theta"]])
elif model == "Bu2019bc":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["phi"],lbols[key]["theta"]])
elif model == "Bu2019re":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["a"],lbols[key]["theta"]])
elif model == "Bu2019op":
param_array.append([np.log10(lbols[key]["kappaLF"]),lbols[key]["gammaLF"],np.log10(lbols[key]["kappaLR"]),lbols[key]["gammaLR"]])
elif model == "Bu2019ops":
param_array.append([np.log10(lbols[key]["kappaLF"]),np.log10(lbols[key]["kappaLR"]),lbols[key]["gammaLR"]])
elif model in ["Bu2019rp","Bu2019rpd"]:
param_array.append([np.log10(lbols[key]["mej_1"]),np.log10(lbols[key]["mej_2"]),lbols[key]["phi"],lbols[key]["a"],lbols[key]["theta"]])
elif model == "Bu2019rps":
param_array.append([np.log10(lbols[key]["mej_1"]),np.log10(lbols[key]["mej_2"]),lbols[key]["a"]])
elif model == "Wo2020dyn":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["a"],lbols[key]["sd"],lbols[key]["theta"]])
elif model == "Wo2020dw":
param_array.append([np.log10(lbols[key]["mej"]),lbols[key]["rwind"],lbols[key]["theta"]])
param_array_postprocess = np.array(param_array)
param_mins, param_maxs = np.min(param_array_postprocess,axis=0),np.max(param_array_postprocess,axis=0)
for i in range(len(param_mins)):
param_array_postprocess[:,i] = (param_array_postprocess[:,i]-param_mins[i])/(param_maxs[i]-param_mins[i])
lbol_array_postprocess = np.array(lbol_array)
mins,maxs = np.min(lbol_array_postprocess,axis=0),np.max(lbol_array_postprocess,axis=0)
for i in range(len(mins)):
lbol_array_postprocess[:,i] = (lbol_array_postprocess[:,i]-mins[i])/(maxs[i]-mins[i])
lbol_array_postprocess[np.isnan(lbol_array_postprocess)]=0.0
UA, sA, VA = np.linalg.svd(lbol_array_postprocess, full_matrices=True)
VA = VA.T
n, n = UA.shape
m, m = VA.shape
cAmat = np.zeros((n_coeff,n))
cAvar = np.zeros((n_coeff,n))
for i in range(n):
cAmat[:,i] = np.dot(lbol_array_postprocess[i,:],VA[:,:n_coeff])
ErrorLevel = 2.0
errors = ErrorLevel*lbol_array_postprocess[i,:]
cAvar[:,i] = np.diag(np.dot(VA[:,:n_coeff].T,np.dot(np.diag(np.power(errors,2.)),VA[:,:n_coeff])))
cAstd = np.sqrt(cAvar)
nsvds, nparams = param_array_postprocess.shape
if gptype == "sklearn":
kernel = 1.0 * RationalQuadratic(length_scale=1.0, alpha=0.1)
gps = []
for i in range(n_coeff):
gp = GaussianProcessRegressor(kernel=kernel,n_restarts_optimizer=0)
gp.fit(param_array_postprocess, cAmat[i,:])
gps.append(gp)
elif gptype == "gpytorch":
# initialize likelihood and model
likelihood = gpytorch.likelihoods.GaussianLikelihood()
#training_iter = 10
training_iter = 1
gps = []
for i in range(n_coeff):
if np.mod(i,5) == 0:
print('Coefficient %d/%d...' % (i, n_coeff))
train_x = torch.from_numpy(param_array_postprocess).float()
train_y = torch.from_numpy(cAmat[i,:]).float()
if torch.cuda.is_available():
train_x = train_x.cuda()
train_y = train_y.cuda()
likelihood = likelihood.cuda()
model = ExactGPModel(train_x, train_y, likelihood)
if torch.cuda.is_available():
model = model.cuda()
# Find optimal model hyperparameters
model.train()
likelihood.train()
# Use the adam optimizer
optimizer = torch.optim.Adam([{'params': model.parameters()}],
lr=0.1)
# "Loss" for GPs - the marginal log likelihood
mll = gpytorch.mlls.ExactMarginalLogLikelihood(likelihood,
model)
for j in range(training_iter):
# Zero gradients from previous iteration
optimizer.zero_grad()
# Output from model
output = model(train_x)
# Calc loss and backprop gradients
loss = -mll(output, train_y)
loss.backward()
#print('Coeff %d/%d Iter %d/%d - Loss: %.3f lengthscale: %.3f noise: %.3f' % (i+1, n_coeff, j + 1, training_iter, loss.item(), model.covar_module.base_kernel.lengthscale.item(), model.likelihood.noise.item()))
optimizer.step()
gps.append(model.state_dict())
svd_model = {}
svd_model["n_coeff"] = n_coeff
svd_model["param_array"] = param_array
svd_model["param_array_postprocess"] = param_array_postprocess
svd_model["cAmat"] = cAmat
svd_model["cAstd"] = cAstd
svd_model["VA"] = VA
svd_model["param_mins"] = param_mins
svd_model["param_maxs"] = param_maxs
svd_model["mins"] = mins
svd_model["maxs"] = maxs
svd_model["gps"] = gps
svd_model["tt"] = tt
print("Finished calculating SVD model of bolometric luminosity...")
return svd_model
