def test_molecules(molecule):
if os.path.isfile('examples/%s.dat' % molecule):
molecule = 'examples/%s' % molecule
elif os.path.isfile('Radex/data/%s.dat' % molecule):
molecule = 'Radex/data/%s' % molecule
else:
return
data = pyradex.pyradex(executable=exepath,species=molecule,minfreq=1,maxfreq=250)
data.pprint(show_unit=True)
def test_molecules(molecule):
if os.path.isfile('examples/%s.dat' % molecule):
molecule = 'examples/%s' % molecule
elif os.path.isfile('Radex/data/%s.dat' % molecule):
molecule = 'Radex/data/%s' % molecule
else:
return
data = pyradex.pyradex(executable=exepath,
species=molecule,
minfreq=1,
maxfreq=250)
data.pprint(show_unit=True)
import timeit
import numpy as np
import textwrap
# Check correctness before doing timing tests
import pyradex
py_pop = [
pyradex.pyradex(collider_densities={
'oH2': 900,
'pH2': 100
},
column=n,
temperature=20)['pop_up'][0]
for n in 10**np.arange(12, 18)
]
R = pyradex.Radex(collider_densities={
'oH2': 900,
'pH2': 100
},
column=1e15,
temperature=20)
R_noreload_pop = []
for n in 10**np.arange(12, 18):
R.column = n
R.run_radex(reload_molfile=False, validate_colliders=False)
R_noreload_pop.append(R.level_population[1])
R_pop = []
def test_call():
data = pyradex.pyradex(executable=exepath,species='Radex/data/hco+',minfreq=50)
data.pprint(show_unit=True)
import timeit
import numpy as np
import textwrap
import warnings
warnings.filterwarnings('ignore')
from astropy import log
log.setLevel(1000)
# Check correctness before doing timing tests
import pyradex
py_pop = [pyradex.pyradex(collider_densities={'oH2':900,'pH2':100},column=n, temperature=20)['pop_up'][0]
for n in 10**np.arange(12,18)]
R = pyradex.Radex(collider_densities={'oH2':900,'pH2':100}, column=1e15, temperature=20)
R_noreload_pop = []
for n in 10**np.arange(12,18):
R.column = n
R.run_radex(reload_molfile=False, validate_colliders=False)
R_noreload_pop.append(R.level_population[1])
R_pop = []
for n in 10**np.arange(12,18):
R.column = n
R.run_radex(reload_molfile=True, validate_colliders=True)
R_pop.append(R.level_population[1])
R_reuse_pop = []
for n in 10**np.arange(12,18):
R.column = n
R.run_radex(reload_molfile=False, validate_colliders=False, reuse_last=True)
def myloglike(cube, ndim, nparams):
import warnings
debug=False # Change this to return likelihoods to diagnose certain problems
# Calculate the log(likelihood). Load in the measdata.pkl for our data,
# and use pyradex to get our model.
# If we violate any priors, limits, or have calculation errors, return -2e100 as the likelihood.
meas=pickle.load(open("measdata.pkl","rb"))
taumin=meas['taulimit'][0]
taumax=meas['taulimit'][1]
sigmacut=meas['sigmacut']
# First, test that the cube parameters do not violate the mass or length priors.
# Warm mass cannot be greater than cold mass.
# Cold temperature cannot be less than warm temperature
# Total mass cannot be greater than dynamical mass limit
# Neither length can be greater than the length limits
if ndim>4:
if cube[6]+cube[7] > cube[2]+cube[3] or \
cube[1] > cube[5] or \
np.log10(np.power(10,cube[2]+cube[3])+np.power(10,cube[6]+cube[7])) > meas['masscut'] or \
cube[2]-cube[0]-0.5*cube[3] > meas['lengthcut'] or \
cube[6]-cube[4]-0.5*cube[7] > meas['lengthcut']:
if debug: return 1e2
return -2e100
else:
if cube[2]+cube[3] > meas['masscut'] or cube[2]-cube[0]-0.5*cube[3] > meas['lengthcut']:
if debug: return 1e2
return -2e100
# Call RADEX for the first component.
try:
dat1=pyradex.pyradex(minfreq=1, maxfreq=1600,
temperature=np.power(10,cube[1]), column=np.power(10,cube[2]),
collider_densities={'H2':np.power(10,cube[0])},
tbg=meas['tbg'], species=meas['head']['mol'], velocity_gradient=1.0, debug=False,
return_dict=True)
# dat['J_up'] returned as strings; this is fine for CO...
jup1=np.array(map(float,dat1['J_up']))
model1=np.array(map(float,dat1['FLUX_Kkms']))*np.power(10,cube[3])
tau1=np.array(map(float,dat1['TAU']))
# Check for convergence
if dat1['niter']==['****']: return -2e100
# At this time it is too slow to use this.
#R = pyradex.Radex(collider_densities={'h2':np.power(10,cube[0])},
# temperature=np.power(10,cube[1]), column=np.power(10,cube[2]),
# tbackground=meas['tbg'],species=meas['head']['mol'],deltav=1.0,debug=False)
#niter=R.run_radex(validate_colliders=False)
#model1=1.064575*R.T_B*np.power(10,cube[3]) # Integrating over velocity, and Filling Factor
#model1=model1.value # Hatred of units in my way
#tau1=R.tau
#jup1=R.upperlevelindex
#cube[ndim]=np.sum(R.source_brightness_beta) # luminosity; not correct units yet.
except:
if debug: return 1e3
return -2e100
# Which indices are lines that we are concerned with?
juse=np.in1d(jup1.ravel(),meas['J_up']).reshape(jup1.shape)
jmeas=np.in1d(jup1.ravel(),meas['J_up'][meas['flux']!=0]).reshape(jup1.shape)
# If applicable, call RADEX for the second component and find their sum.
# Either way, check if any optical depths are outside of our limits.
if ndim>4:
try:
dat2=pyradex.pyradex(minfreq=1, maxfreq=1600,
temperature=np.power(10,cube[5]), column=np.power(10,cube[6]),
collider_densities={'H2':np.power(10,cube[4])},
tbg=meas['tbg'], species=meas['head']['mol'], velocity_gradient=1.0, debug=False,
return_dict=True)
jup2=np.array(map(float,dat2['J_up']))
model2=np.array(map(float,dat2['FLUX_Kkms']))*np.power(10,cube[7])
tau2=np.array(map(float,dat2['TAU']))
# Check for convergence
if dat2['niter']==['****']: return -2e100
modelt=model1+model2# We want to compare the SUM of the components to our data.
tauok=np.all([tau1<taumax,tau1>taumin,tau2<taumax,tau2>taumin],axis=0)
#R.temperature=np.power(10,cube[5])
#R.density=np.power(10,cube[4])
#R.column=np.power(10,cube[6])
#niter=R.run_radex(validate_colliders=False)
#model2=1.064575*R.T_B*np.power(10,cube[3])# Integrating over velocity, and Filling Factor
#model2=model2.value
#tau2=R.tau
#jup2=R.upperlevelindex
#if not np.array_equal(jup1,jup2):
# warnings.warn('J_up arrays NOT equal from multiple RADEX calls!')
# return -2e100
#modelt=model1+model2
#tauok=np.all([tau1<taumax,tau1>taumin,tau2<taumax,tau2>taumin],axis=0)
except:
if debug: return 1e3
return -2e100
else:
modelt=model1# total model
tauok=np.all([tau1<taumax,tau1>taumin],axis=0)
# Check that we have at least one line with optical depth in allowable limits.
if not np.any(tauok[jmeas]):
if debug: return 1e4
return -2e100
# Check that we don't violate ANY line flux upper limits.
ul=np.where(meas['flux']==0)[0]
for i in ul:
ulok=modelt[jup1==meas['J_up'][i]] < meas['sigma'][i]*sigmacut
if not ulok:
if debug: return 1e5
return -2e100
# We've made it! Calculate likelihood!
nmeas=len(meas['sigma'])
loglike=-nmeas*0.5*np.log(2.0*np.pi)
for i in range(nmeas):
try:
j_ind=np.where(jup1==meas['J_up'][i])[0]
if tauok[j_ind] and meas['flux'][i] > 0:
loglike=loglike-np.log(meas['sigma'][i])
loglike=loglike-0.5*(np.power((meas['flux'][i]-modelt[j_ind])/meas['sigma'][i],2))
except:
warnings.warn('An error occured with j_up = '+str(meas['J_up'][i])+', will be ignored.')
loglike=loglike
# Record the luminosity, pressure, and beam-averaged column density for binning later.
# The public distribution of RADEX will not output this luminosity; sorry all users...
if 'LogLmol' in dat1.keys():
cube[ndim]=dat1['LogLmol']+meas['areacm']+cube[2]+cube[3]
else:
cube[ndim]=1.0 # Those not using our private RADEX code will not get luminosity.
cube[ndim+1]=cube[0]+cube[1]
cube[ndim+2]=cube[2]+cube[3]
# Don't include sled likelihoods with bad tau - cannot NaN them :( ?
# Watch out, in the rare chance of a ridiculous RADEX value of something E+99,
# MultiNest will print 0.XYZ+100 instead of X.YZE+99, and will not be read as float.
# You've not used this value in your likelihood because tau will be wild as well.
model1[model1>9e98]=9e98
if ndim>4: model2[model2>9e98]=9e98
# If we have 2 components, also records those L, P, and BACD, as well
# as ratios of the warm to cold (note these are in log, so they are differenced).
if ndim>4:
if 'LogLmol' in dat2.keys():
cube[ndim+3]=dat2['LogLmol']+meas['areacm']+cube[6]+cube[7]
else:
cube[ndim+3]=1.0 # See luminosity note above.
cube[ndim+4]=cube[4]+cube[5]
cube[ndim+5]=cube[6]+cube[7]
# Ratios of WARM to COLD
cube[ndim+6]=cube[ndim+3]-cube[ndim]
cube[ndim+7]=cube[ndim+4]-cube[ndim+1]
cube[ndim+8]=cube[ndim+5]-cube[ndim+2]
# SLED likelihoods, unhappy with slice indices...
for i,l in enumerate(model1[0:meas['sled_to_j']]): cube[ndim+9+i]=l
for i,l in enumerate(model2[0:meas['sled_to_j']]): cube[ndim+9+meas['sled_to_j']+i]=l
#cube[ndim+9:]=np.append(model1[0:meas['sled_to_j']].value,model2[0:meas['sled_to_j']])
else:
for i,l in enumerate(model1[0:meas['sled_to_j']]): cube[ndim+3+i]=l
#cube[ndim+3:ndim+3+meas['sled_to_j']]=model1[0:meas['sled_to_j']].value
return loglike
def test_call():
data = pyradex.pyradex(executable=exepath,
species='Radex/data/hco+',
minfreq=50)
data.pprint(show_unit=True)
def myloglike(cube, ndim, nparams):
import warnings
debug = False # Change this to return likelihoods to diagnose certain problems
# Calculate the log(likelihood). Load in the measdata.pkl for our data,
# and use pyradex to get our model.
# If we violate any priors, limits, or have calculation errors, return -2e100 as the likelihood.
meas = pickle.load(open("measdata.pkl", "rb"))
taumin = meas['taulimit'][0]
taumax = meas['taulimit'][1]
sigmacut = meas['sigmacut']
# First, test that the cube parameters do not violate the mass or length priors.
# Warm mass cannot be greater than cold mass.
# Cold temperature cannot be less than warm temperature
# Total mass cannot be greater than dynamical mass limit
# Neither length can be greater than the length limits
if ndim > 4:
if cube[6]+cube[7] > cube[2]+cube[3] or \
cube[1] > cube[5] or \
np.log10(np.power(10,cube[2]+cube[3])+np.power(10,cube[6]+cube[7])) > meas['masscut'] or \
cube[2]-cube[0]-0.5*cube[3] > meas['lengthcut'] or \
cube[6]-cube[4]-0.5*cube[7] > meas['lengthcut']:
if debug: return 1e2
return -2e100
else:
if cube[2] + cube[3] > meas['masscut'] or cube[2] - cube[
0] - 0.5 * cube[3] > meas['lengthcut']:
if debug: return 1e2
return -2e100
# Call RADEX for the first component.
try:
dat1 = pyradex.pyradex(
minfreq=1,
maxfreq=1600,
temperature=np.power(10, cube[1]),
column=np.power(10, cube[2]),
collider_densities={'H2': np.power(10, cube[0])},
tbg=meas['tbg'],
species=meas['head']['mol'],
velocity_gradient=1.0,
debug=False,
return_dict=True)
# dat['J_up'] returned as strings; this is fine for CO...
jup1 = np.array(map(float, dat1['J_up']))
model1 = np.array(map(float, dat1['FLUX_Kkms'])) * np.power(
10, cube[3])
tau1 = np.array(map(float, dat1['TAU']))
# Check for convergence
if dat1['niter'] == ['****']: return -2e100
# At this time it is too slow to use this.
#R = pyradex.Radex(collider_densities={'h2':np.power(10,cube[0])},
# temperature=np.power(10,cube[1]), column=np.power(10,cube[2]),
# tbackground=meas['tbg'],species=meas['head']['mol'],deltav=1.0,debug=False)
#niter=R.run_radex(validate_colliders=False)
#model1=1.064575*R.T_B*np.power(10,cube[3]) # Integrating over velocity, and Filling Factor
#model1=model1.value # Hatred of units in my way
#tau1=R.tau
#jup1=R.upperlevelindex
#cube[ndim]=np.sum(R.source_brightness_beta) # luminosity; not correct units yet.
except:
if debug: return 1e3
return -2e100
# Which indices are lines that we are concerned with?
juse = np.in1d(jup1.ravel(), meas['J_up']).reshape(jup1.shape)
jmeas = np.in1d(jup1.ravel(),
meas['J_up'][meas['flux'] != 0]).reshape(jup1.shape)
# If applicable, call RADEX for the second component and find their sum.
# Either way, check if any optical depths are outside of our limits.
if ndim > 4:
try:
dat2 = pyradex.pyradex(
minfreq=1,
maxfreq=1600,
temperature=np.power(10, cube[5]),
column=np.power(10, cube[6]),
collider_densities={'H2': np.power(10, cube[4])},
tbg=meas['tbg'],
species=meas['head']['mol'],
velocity_gradient=1.0,
debug=False,
return_dict=True)
jup2 = np.array(map(float, dat2['J_up']))
model2 = np.array(map(float, dat2['FLUX_Kkms'])) * np.power(
10, cube[7])
tau2 = np.array(map(float, dat2['TAU']))
# Check for convergence
if dat2['niter'] == ['****']: return -2e100
modelt = model1 + model2 # We want to compare the SUM of the components to our data.
tauok = np.all(
[tau1 < taumax, tau1 > taumin, tau2 < taumax, tau2 > taumin],
axis=0)
#R.temperature=np.power(10,cube[5])
#R.density=np.power(10,cube[4])
#R.column=np.power(10,cube[6])
#niter=R.run_radex(validate_colliders=False)
#model2=1.064575*R.T_B*np.power(10,cube[3])# Integrating over velocity, and Filling Factor
#model2=model2.value
#tau2=R.tau
#jup2=R.upperlevelindex
#if not np.array_equal(jup1,jup2):
# warnings.warn('J_up arrays NOT equal from multiple RADEX calls!')
# return -2e100
#modelt=model1+model2
#tauok=np.all([tau1<taumax,tau1>taumin,tau2<taumax,tau2>taumin],axis=0)
except:
if debug: return 1e3
return -2e100
else:
modelt = model1 # total model
tauok = np.all([tau1 < taumax, tau1 > taumin], axis=0)
# Check that we have at least one line with optical depth in allowable limits.
if not np.any(tauok[jmeas]):
if debug: return 1e4
return -2e100
# Check that we don't violate ANY line flux upper limits.
ul = np.where(meas['flux'] == 0)[0]
for i in ul:
ulok = modelt[jup1 == meas['J_up'][i]] < meas['sigma'][i] * sigmacut
if not ulok:
if debug: return 1e5
return -2e100
# We've made it! Calculate likelihood!
nmeas = len(meas['sigma'])
loglike = -nmeas * 0.5 * np.log(2.0 * np.pi)
for i in range(nmeas):
try:
j_ind = np.where(jup1 == meas['J_up'][i])[0]
if tauok[j_ind] and meas['flux'][i] > 0:
loglike = loglike - np.log(meas['sigma'][i])
loglike = loglike - 0.5 * (np.power(
(meas['flux'][i] - modelt[j_ind]) / meas['sigma'][i], 2))
except:
warnings.warn('An error occured with j_up = ' +
str(meas['J_up'][i]) + ', will be ignored.')
loglike = loglike
# Record the luminosity, pressure, and beam-averaged column density for binning later.
# The public distribution of RADEX will not output this luminosity; sorry all users...
if 'LogLmol' in dat1.keys():
cube[ndim] = dat1['LogLmol'] + meas['areacm'] + cube[2] + cube[3]
else:
cube[
ndim] = 1.0 # Those not using our private RADEX code will not get luminosity.
cube[ndim + 1] = cube[0] + cube[1]
cube[ndim + 2] = cube[2] + cube[3]
# Don't include sled likelihoods with bad tau - cannot NaN them :( ?
# Watch out, in the rare chance of a ridiculous RADEX value of something E+99,
# MultiNest will print 0.XYZ+100 instead of X.YZE+99, and will not be read as float.
# You've not used this value in your likelihood because tau will be wild as well.
model1[model1 > 9e98] = 9e98
if ndim > 4: model2[model2 > 9e98] = 9e98
# If we have 2 components, also records those L, P, and BACD, as well
# as ratios of the warm to cold (note these are in log, so they are differenced).
if ndim > 4:
if 'LogLmol' in dat2.keys():
cube[ndim +
3] = dat2['LogLmol'] + meas['areacm'] + cube[6] + cube[7]
else:
cube[ndim + 3] = 1.0 # See luminosity note above.
cube[ndim + 4] = cube[4] + cube[5]
cube[ndim + 5] = cube[6] + cube[7]
# Ratios of WARM to COLD
cube[ndim + 6] = cube[ndim + 3] - cube[ndim]
cube[ndim + 7] = cube[ndim + 4] - cube[ndim + 1]
cube[ndim + 8] = cube[ndim + 5] - cube[ndim + 2]
# SLED likelihoods, unhappy with slice indices...
for i, l in enumerate(model1[0:meas['sled_to_j']]):
cube[ndim + 9 + i] = l
for i, l in enumerate(model2[0:meas['sled_to_j']]):
cube[ndim + 9 + meas['sled_to_j'] + i] = l
#cube[ndim+9:]=np.append(model1[0:meas['sled_to_j']].value,model2[0:meas['sled_to_j']])
else:
for i, l in enumerate(model1[0:meas['sled_to_j']]):
cube[ndim + 3 + i] = l
#cube[ndim+3:ndim+3+meas['sled_to_j']]=model1[0:meas['sled_to_j']].value
return loglike
def myloglike(cube, ndim, nparams):
# Calculate the log(likelihood). Load in the measdata.pkl for our data,
# and use pyradex to get our model.
# If we violate any priors, limits, or have calculation errors, return -2e100 as the likelihood.
meas=pickle.load(open("measdata.pkl","rb"))
taumin=meas['taulimit'][0]#-0.9
taumax=meas['taulimit'][1]#100.0
sigmacut=meas['sigmacut']#3.0
import warnings
# First, test that the cube parameters do not violate the mass or length priors.
if ndim>4:
if cube[6]+cube[7] > cube[2]+cube[3] or \
cube[1] > cube[5] or \
math.log10(math.pow(10,cube[2]+cube[3])+math.pow(10,cube[6]+cube[7])) > meas['masscut'] or \
cube[2]-cube[0]-0.5*cube[3] > meas['lengthcut'] or \
cube[6]-cube[4]-0.5*cube[7] > meas['lengthcut']:
return -2e100
else:
if cube[2]+cube[3] > meas['masscut'] or cube[2]-cube[0]-0.5*cube[3] > meas['lengthcut']:
return -2e100
# Call RADEX for the first component.
try:
dat=pyradex.pyradex(minfreq=1, maxfreq=1600,
temperature=math.pow(10,cube[1]), column=math.pow(10,cube[2]),
collider_densities={'H2':math.pow(10,cube[0])},
tbg=2.73, species='co', velocity_gradient=1.0, debug=True)
dat['FLUX_Kkms']*=math.pow(10,cube[3])
except:
return -2e100
# If applicable, call RADEX for the second component and find their sum.
# Either way, check if any optical depths are outside of our limits.
if ndim>4:
try:
dat2=pyradex.pyradex(minfreq=1, maxfreq=1600,
temperature=math.pow(10,cube[5]), column=math.pow(10,cube[6]),
collider_densities={'H2':math.pow(10,cube[4])},
tbg=2.73, species='co', velocity_gradient=1.0, debug=False)
dat2['FLUX_Kkms']*=math.pow(10,cube[7])
newdat=dat['FLUX_Kkms']+dat2['FLUX_Kkms'] # We want to compare the SUM of the components to our data.
tauok=numpy.all([dat['TAU']<taumax,dat['TAU']>taumin,dat2['TAU']<taumax,dat2['TAU']>taumin],axis=0)
except:
return -2e100
else:
newdat=dat['FLUX_Kkms']
tauok=numpy.all([dat['TAU']<taumax,dat['TAU']>taumin],axis=0)
# Check that we have at least one line with optical depth in allowable limits.
if not numpy.any(tauok):
return -2e100
# Check that we don't violate ANY line flux upper limits.
ulok=newdat[meas['flux']==0] < meas['sigma'][meas['flux']==0]*sigmacut
if not numpy.all(ulok):
return -2e100
# We've made it! Calculate likelihood!
nmeas=len(meas['sigma'])
loglike=-nmeas*0.5*math.log(2.0*math.pi)
for i in range(nmeas):
try:
if tauok[i] and meas['flux'][i] > 0:
loglike=loglike-math.log(meas['sigma'][i])
loglike=loglike-0.5*(math.pow((meas['flux'][i]-newdat[dat['J_up'] == meas['J_up'][i]])/meas['sigma'][i],2))
except:
warnings.warn('An error occured with J_up = '+str(meas['J_up'][i])+', will be ignored.')
loglike=loglike
# Record the luminosity, pressure, and beam-averaged column density for binning later.
cube[ndim]=dat['LogLmol']+meas['areacm']+cube[2]+cube[3] #0
cube[ndim+1]=cube[0]+cube[1]
cube[ndim+2]=cube[2]+cube[3]
# If we have 2 components, also records those L, P, and BACD, as well
# as ratios of the warm to cold (note these are in log, so they are differenced).
if ndim>4:
cube[ndim+3]=dat2['LogLmol']+meas['areacm']+cube[6]+cube[7] #0
cube[ndim+4]=cube[4]+cube[5]
cube[ndim+5]=cube[6]+cube[7]
# Ratios of WARM to COLD
cube[ndim+6]=cube[ndim+3]-cube[ndim]
cube[ndim+7]=cube[ndim+4]-cube[ndim+1]
cube[ndim+8]=cube[ndim+5]-cube[ndim+2]
return loglike
