def initialize_model(planets, TESSAPFdata, addextra=False):
time_base = 2458362.5
nplan = len(planets)
if addextra:
nplan = nplan + 1
params = radvel.Parameters(
nplan, basis='per tp e w k') # number of planets = nplan
n = 1
for idx in planets:
c_period = TESSAPFdata['period'][idx]
k = radvel.utils.semi_amplitude(TESSAPFdata['est_mass'][idx], c_period,
TESSAPFdata['mstar'][idx], 0.)
tpstart = time_base + c_period * (TESSAPFdata['phase'][idx] + 0.25)
params['per%d' % (n)] = radvel.Parameter(value=c_period)
params['tp%d' % (n)] = radvel.Parameter(value=tpstart)
params['e%d' % (n)] = radvel.Parameter(value=0.0001)
params['w%d' % (n)] = radvel.Parameter(value=np.pi / 2.)
params['k%d' % (n)] = radvel.Parameter(value=k)
n = n + 1
if addextra:
params['per%d' % (n)] = radvel.Parameter(value=30.)
params['tp%d' % (n)] = radvel.Parameter(value=time_base + 15.)
params['e%d' % (n)] = radvel.Parameter(value=0.0001)
params['w%d' % (n)] = radvel.Parameter(value=np.pi / 2.)
params['k%d' % (n)] = radvel.Parameter(value=2.)
mod = radvel.RVModel(params, time_base=time_base)
mod.params['dvdt'] = radvel.Parameter(value=0.0)
mod.params['curv'] = radvel.Parameter(value=0.0)
return mod
def initialize_post(data, params=None, priors=[], linear=True, decorrs=None):
"""Initialize a posterior object with data, params, and priors.
Args:
data: a pandas dataframe.
params: a list of radvel parameter objects.
priors: a list of priors to place on the posterior object.
decorrs: a list of decorrelation vectors.
Returns:
post (radvel Posterior object)
"""
if params is None:
# params = radvel.Parameters(1, basis='per tc secosw sesinw logk')
params = initialize_default_pars(instnames=data.tel, times=data.time)
iparams = radvel.basis._copy_params(params)
# Allow for time to be listed as 'time' or 'jd' (Julian Date).
if {'jd'}.issubset(data.columns):
data['time'] = data['jd']
# initialize RVModel
time_base = np.mean([data['time'].max(), data['time'].min()])
mod = radvel.RVModel(params, time_base=time_base)
# initialize Likelihood objects for each instrument
telgrps = data.groupby('tel').groups
likes = {}
for inst in telgrps.keys():
# 10/8: ADD DECORRELATION VECTORS AND VARS, ONLY FOR SELECTED INST.
likes[inst] = radvel.likelihood.RVLikelihood(
mod,
data.iloc[telgrps[inst]].time,
data.iloc[telgrps[inst]].mnvel,
data.iloc[telgrps[inst]].errvel,
suffix='_' + inst)
likes[inst].params['gamma_' + inst] = iparams['gamma_' + inst]
likes[inst].params['jit_' + inst] = iparams['jit_' + inst]
# Can this be cleaner? like = radvel.likelihood.CompositeLikelihood(likes)
like = radvel.likelihood.CompositeLikelihood(list(likes.values()))
post = radvel.posterior.Posterior(like)
if priors == []:
priors.append(radvel.prior.PositiveKPrior(post.params.num_planets))
priors.append(radvel.prior.EccentricityPrior(post.params.num_planets))
if not linear:
if ('j' in telgrps.keys()) and ('k' in telgrps.keys()):
TexStr = 'Gaussian Prior on HIRES offset'
OffsetPrior = radvel.prior.UserDefinedPrior(
['gamma_j', 'gamma_k'], GaussianDiffFunc, TexStr)
priors.append(OffsetPrior)
#for inst in telgrps.keys():
# priors.append(radvel.prior.Jeffrey('jit_'+inst, 0.01, 20.0))
post.priors = priors
return post
def initialize_model(e):
time_base = 0.0
params = radvel.Parameters(1, basis='per tp e w k')
params['per1'] = radvel.Parameter(value=6)
params['tp1'] = radvel.Parameter(value=2)
params['e1'] = radvel.Parameter(value=e)
params['w1'] = radvel.Parameter(value=0.0)
params['k1'] = radvel.Parameter(value=10.1)
mod = radvel.RVModel(params, time_base=time_base)
mod.params['dvdt'] = radvel.Parameter(value=-0.0)
mod.params['curv'] = radvel.Parameter(value=0.0)
return mod
def initialize_posterior(P):
params = P.params.basis.from_cps(P.params, P.fitting_basis, keep=False)
for key in params.keys():
if key.startswith('logjit'):
msg = """
Fitting log(jitter) is depreciated. Please convert your config
files to initialize 'jit' instead of 'logjit' parameters.
Converting 'logjit' to 'jit' for you now.
"""
warnings.warn(msg, DeprecationWarning, stacklevel=2)
newkey = key.replace('logjit', 'jit')
params[newkey] = np.exp(params[key])
P.vary[newkey] = P.vary[key]
del P.vary[key]
del params[key]
iparams = params.copy()
# Make sure we don't have duplicate indicies in the DataFrame
P.data = P.data.reset_index(drop=True)
# initialize RVmodel object
mod = radvel.RVModel(params, time_base=P.time_base)
# initialize RVlikelihood objects for each instrument
telgrps = P.data.groupby('tel').groups
likes = {}
for inst in P.instnames:
likes[inst] = radvel.likelihood.RVLikelihood(
mod,
P.data.iloc[telgrps[inst]].time,
P.data.iloc[telgrps[inst]].mnvel,
P.data.iloc[telgrps[inst]].errvel,
suffix='_' + inst)
likes[inst].params['gamma_' + inst] = iparams['gamma_' + inst]
likes[inst].params['jit_' + inst] = iparams['jit_' + inst]
like = radvel.likelihood.CompositeLikelihood(likes.values())
# Set fixed/vary parameters
like.vary.update(P.vary)
# Initialize Posterior object
post = radvel.posterior.Posterior(like)
post.priors = P.priors
return post
def initialize_model(planets, TESSAPFdata, addextra=False):
time_base = 2458362.5
nplan = len(planets)
if addextra:
nplan = nplan + 1
params = radvel.Parameters(
nplan, basis='per tp e w k') # number of planets = nplan
n = 1
for idx in planets:
c_period = TESSAPFdata['period'][idx]
k = radvel.utils.semi_amplitude(TESSAPFdata['est_mass'][idx], c_period,
TESSAPFdata['mstar'][idx], 0.)
tpstart = time_base + c_period * (TESSAPFdata['phase'][idx] + 0.25)
params['per%d' % (n)] = radvel.Parameter(value=c_period)
params['tp%d' % (n)] = radvel.Parameter(value=tpstart)
params['e%d' % (n)] = radvel.Parameter(value=0.0)
params['w%d' % (n)] = radvel.Parameter(value=0.)
params['k%d' % (n)] = radvel.Parameter(value=k)
n = n + 1
if addextra:
params['per%d' % (n)] = radvel.Parameter(value=30.)
params['tp%d' % (n)] = radvel.Parameter(value=time_base + 15.)
params['e%d' % (n)] = radvel.Parameter(value=0.0001)
params['w%d' % (n)] = radvel.Parameter(value=np.pi / 2.)
params['k%d' % (n)] = radvel.Parameter(value=2.)
# nparams = params.basis.from_cps(params,'per tc e w k') # transform to new basis (for less biased fits?)
# n = 1
# if addextra:
# n = 2
# for i in range(1,len(planets+n)):
# # del nparams['e%d' % (i)]
# # # del nparams['k%d' % (i)]
# # del nparams['w%d' % (i)]
# del nparams['tp%d' % (i)]
# # new basis
mod = radvel.RVModel(params, time_base=time_base)
mod.params['dvdt'] = radvel.Parameter(value=0.0)
mod.params['curv'] = radvel.Parameter(value=0.0)
return mod
def get_full_like(observations_df):
"""
Get a radvel likelihood object for a two-planet radial velocity model.
Arguments
---------
observations_df : pandas DataFrame
Dataframe containing the radial observations.
Required columns:
'instrument','time','velocity','uncertainty'
Returns
-------
radvel CompositeLikelihood
"""
# Set up radvel likelihood object
fitbasis = 'per tc e w k'
params = radvel.Parameters(2, basis=fitbasis)
mdl = radvel.RVModel(params, time_base=observations_df.time.median())
likes = []
instruments = observations_df.instrument.unique()
for i, instrument in enumerate(instruments):
data = observations_df.query('instrument==@instrument')[[
'time', 'velocity', 'uncertainty'
]].values
inst_like = radvel.likelihood.RVLikelihood(mdl,
*(data.T),
suffix=instrument)
# ... setting pars to 0 is for some reason
# neccessary to stop the code from eating shit...
for key, par in inst_like.params.items():
par.value = 0
likes.append(inst_like)
like = radvel.likelihood.CompositeLikelihood(likes)
like.params['dvdt'].vary = False
like.params['curv'].vary = False
return like
def initialize_model(self, periods, numplanets):
"""
Initialize the RadVel keplerian analysis model as described in the tutorial for each of the n number of planets
specified Fulton et al. (2018).
:param periods: The periods specified for each planet.
:param numplanets: The number of planets in the system.
:return: mod, the initialized radvel model.
"""
time_base = self.t.values[0]
params = radvel.Parameters(numplanets,
basis='per tc secosw sesinw logk')
for i in range(1, numplanets + 1):
params['per' + str(i)] = radvel.Parameter(value=periods[i - 1])
params['tc' + str(i)] = radvel.Parameter(value=self.t[0])
params['secosw' + str(i)] = radvel.Parameter(value=0.01)
params['sesinw' + str(i)] = radvel.Parameter(value=0.01)
params['logk' + str(i)] = radvel.Parameter(value=1.1)
mod = radvel.RVModel(params, time_base=time_base)
mod.params['dvdt'] = radvel.Parameter(value=-0.02)
mod.params['curv'] = radvel.Parameter(value=0.01)
print(mod)
return mod
def initialize_posterior(config_file, decorr=False):
"""Initialize Posterior object
Parse a setup file and initialize the RVModel, Likelihood, Posterior and priors.
Args:
config_file (string): path to config file
decorr (bool): (optional) decorrelate RVs against columns defined in the decorr_vars list
Returns:
tuple: (object representation of config file, radvel.Posterior object)
"""
system_name = os.path.basename(config_file).split('.')[0]
P = imp.load_source(system_name, os.path.abspath(config_file))
params = P.params
assert str(params.basis) == "Basis Object <{}>".format(
P.fitting_basis), """
Parameters in config file must be converted to fitting basis.
"""
if decorr:
try:
decorr_vars = P.decorr_vars
except:
raise Exception("--decorr option selected,\
but decorr_vars is not found in your setup file.")
else:
decorr_vars = []
for key in params.keys():
if key.startswith('logjit'):
msg = """
Fitting log(jitter) is depreciated. Please convert your config
files to initialize 'jit' instead of 'logjit' parameters.
Converting 'logjit' to 'jit' for you now.
"""
warnings.warn(msg, DeprecationWarning, stacklevel=2)
newkey = key.replace('logjit', 'jit')
params[newkey] = radvel.model.Parameter(value=np.exp(
params[key].value),
vary=params[key].vary)
del params[key]
iparams = radvel.basis._copy_params(params)
# Make sure we don't have duplicate indicies in the DataFrame
P.data = P.data.reset_index(drop=True)
# initialize RVmodel object
mod = radvel.RVModel(params, time_base=P.time_base)
# initialize Likelihood objects for each instrument
telgrps = P.data.groupby('tel').groups
likes = {}
for inst in P.instnames:
assert inst in P.data.groupby('tel').groups.keys(), \
"No data found for instrument '{}'.\nInstruments found in this dataset: {}".format(inst,
list(telgrps.keys()))
decorr_vectors = {}
if decorr:
for d in decorr_vars:
decorr_vectors[d] = P.data.iloc[telgrps[inst]][d].values
try:
hnames = P.hnames[inst]
liketype = radvel.likelihood.GPLikelihood
try:
kernel_name = P.kernel_name[inst]
if kernel_name == "Celerite":
liketype = radvel.likelihood.CeleriteLikelihood
except AttributeError:
kernel_name = "QuasiPer"
except AttributeError:
liketype = radvel.likelihood.RVLikelihood
kernel_name = None
hnames = None
likes[inst] = liketype(mod,
P.data.iloc[telgrps[inst]].time,
P.data.iloc[telgrps[inst]].mnvel,
P.data.iloc[telgrps[inst]].errvel,
hnames=hnames,
suffix='_' + inst,
kernel_name=kernel_name,
decorr_vars=decorr_vars,
decorr_vectors=decorr_vectors)
likes[inst].params['gamma_' + inst] = iparams['gamma_' + inst]
likes[inst].params['jit_' + inst] = iparams['jit_' + inst]
like = radvel.likelihood.CompositeLikelihood(list(likes.values()))
# Initialize Posterior object
post = radvel.posterior.Posterior(like)
post.priors = P.priors
return P, post
bin_t, bin_vel, bin_err, bin_tel = radvel.utils.bintels(data['time'].values, data['mnvel'].values, data['errvel'].values, data['tel'].values, binsize=0.1) data = pd.DataFrame([], columns=['time', 'mnvel', 'errvel', 'tel']) data['time'] = bin_t data['mnvel'] = bin_vel data['errvel'] = bin_err data['tel'] = bin_tel instnames = ['j', 'k'] ntels = len(instnames) anybasis_params['gamma_j'] = radvel.Parameter(value=0.0, vary=False, linear=True) anybasis_params['jit_j'] = radvel.Parameter(value=1.0) anybasis_params['gamma_k'] = radvel.Parameter(value=0.0, vary=False, linear=True) anybasis_params['jit_k'] = radvel.Parameter(value=1.0) params = anybasis_params.basis.to_any_basis(anybasis_params,fitting_basis) mod = radvel.RVModel(params, time_base=time_base) mod.params['per1'].vary = True mod.params['tc1'].vary = True mod.params['secosw1'].vary = True mod.params['sesinw1'].vary = True mod.params['per2'].vary = True mod.params['tc2'].vary = True mod.params['secosw2'].vary = True mod.params['sesinw2'].vary = True mod.params['dvdt'].vary = True mod.params['curv'].vary = False mod.params['jit_j'].vary = True mod.params['jit_k'].vary = True priors = [
if '-fname' in sys.argv:
p = sys.argv.index('-fname')
fname = str(sys.argv[p + 1])
# generate the data
np.random.seed(2021) #For reproducibility
x_rv = np.sort(np.random.uniform(low=1, high=100, size=25))
yerr_rv = sigma
synth_params = radvel.Parameters(1, basis='per tc e w k')
synth_params['per1'] = radvel.Parameter(value=per)
synth_params['tc1'] = radvel.Parameter(value=tc)
synth_params['e1'] = radvel.Parameter(value=e)
synth_params['w1'] = radvel.Parameter(value=0.0)
synth_params['k1'] = radvel.Parameter(value=k)
synth_params['dvdt'] = radvel.Parameter(value=0)
synth_params['curv'] = radvel.Parameter(value=0)
synth_model = radvel.RVModel(params=synth_params)
y_rv = synth_model(x_rv)
np.random.seed(2021) #For reproducibility
y_rv += yerr_rv * np.random.randn(len(y_rv))
data = pd.DataFrame({'time': x_rv, 'mnvel': y_rv, 'errvel': yerr_rv})
data.to_csv(f"./data/{fname}.csv", index=False)
def initialize_sim_posterior(data, mass_c, sma_c, incl_c, ecc_c,
gammajit_dict, verbose=True):
"""
Initialize Posterior object to be used for the "second planet" models.
Basically a hack of radvel.utils.initialize_posterior
Returns:
tuple: (object representation of config file, radvel.Posterior object)
"""
system_name = 'simWASP4'
mstar = 0.864*u.Msun
mass_b = 1.186*u.Mjup
Mtotal = (mass_c + mass_b + mstar).to(u.Msun).value
Msini = (mass_c * np.sin(np.deg2rad(incl_c))).to(u.Mjup).value
period_c = np.sqrt(
4*np.pi**2 * sma_c**3 / (const.G * Mtotal*u.Msun)
).to(u.day).value
k_c = semi_amplitude(Msini, period_c, Mtotal, ecc_c, Msini_units='jupiter')
P = templateWASP4(data, period_c, ecc_c, k_c, tc_c=2455470,
w_c=0.42*np.pi/2, gammajit_dict=gammajit_dict)
# initalization from radvel.utils.initialize_posterior
params = P.params
assert str(params.basis) == "Basis Object <{}>".format(P.fitting_basis), """
Parameters in config file must be converted to fitting basis.
"""
decorr = False
decorr_vars = []
for key in params.keys():
if key.startswith('logjit'):
msg = """
Fitting log(jitter) is depreciated. Please convert your config
files to initialize 'jit' instead of 'logjit' parameters.
Converting 'logjit' to 'jit' for you now.
"""
warnings.warn(msg, DeprecationWarning, stacklevel=2)
newkey = key.replace('logjit', 'jit')
params[newkey] = (
radvel.model.Parameter(value=np.exp(params[key].value),
vary=params[key].vary)
)
del params[key]
iparams = radvel.basis._copy_params(params)
# Make sure we don't have duplicate indicies in the DataFrame
P.data = P.data.reset_index(drop=True)
# initialize RVmodel object
mod = radvel.RVModel(params, time_base=P.time_base)
# initialize Likelihood objects for each instrument
telgrps = P.data.groupby('tel').groups
likes = {}
for inst in P.instnames:
assert inst in P.data.groupby('tel').groups.keys(), \
"No data found for instrument '{}'.\nInstruments found in this dataset: {}".format(inst, list(telgrps.keys()))
decorr_vectors = {}
if decorr:
for d in decorr_vars:
decorr_vectors[d] = P.data.iloc[telgrps[inst]][d].values
try:
hnames = P.hnames[inst]
liketype = radvel.likelihood.GPLikelihood
try:
kernel_name = P.kernel_name[inst]
# if kernel_name == "Celerite":
# liketype = radvel.likelihood.CeleriteLikelihood
if kernel_name == "Celerite":
liketype = radvel.likelihood.CeleriteLikelihood
except AttributeError:
kernel_name = "QuasiPer"
except AttributeError:
liketype = radvel.likelihood.RVLikelihood
kernel_name = None
hnames = None
likes[inst] = liketype(
mod, P.data.iloc[telgrps[inst]].time,
P.data.iloc[telgrps[inst]].mnvel,
P.data.iloc[telgrps[inst]].errvel, hnames=hnames, suffix='_'+inst,
kernel_name=kernel_name, decorr_vars=decorr_vars,
decorr_vectors=decorr_vectors
)
likes[inst].params['gamma_'+inst] = iparams['gamma_'+inst]
likes[inst].params['jit_'+inst] = iparams['jit_'+inst]
like = radvel.likelihood.CompositeLikelihood(list(likes.values()))
# Initialize Posterior object
post = radvel.posterior.Posterior(like)
post.priors = P.priors
#return P, post
return post
params['per2'] = radvel.Parameter(value=Ps[1])
params['tc2'] = radvel.Parameter(value=2458415.6344)
params['secosw2'] = radvel.Parameter(value=np.sqrt(es[1]) * np.cos(omegas[1]))
params['sesinw2'] = radvel.Parameter(value=np.sqrt(es[1]) * np.sin(omegas[1]))
params['k2'] = radvel.Parameter(value=Ks[1])
params['per3'] = radvel.Parameter(value=Ps[2])
params['tc3'] = radvel.Parameter(value=2458409.7328)
params['secosw3'] = radvel.Parameter(value=np.sqrt(es[2]) * np.cos(omegas[2]))
params['sesinw3'] = radvel.Parameter(value=np.sqrt(es[2]) * np.sin(omegas[2]))
params['k3'] = radvel.Parameter(value=Ks[2])
# params['dvdt'] = radvel.Parameter(value=0)
# params['curv'] = radvel.Parameter(value=0)
rv_mod = radvel.RVModel(params)
like_syn = radvel.likelihood.RVLikelihood(rv_mod, t, syn_rv,
np.zeros(t.size) + errval)
like_syn.params['gamma'] = radvel.Parameter(value=0)
like_syn.params['jit'] = radvel.Parameter(value=jitter)
like_syn.params['jit'].vary = False # Don't vary jitter
# like_syn.params['k1'].vary = False # Don't vary period
# like_syn.params['k2'].vary = False # Don't vary period
# like_syn.params['k3'].vary = False # Don't vary period
# like_syn.params['per1'].vary = False # Don't vary period
# like_syn.params['per2'].vary = False # Don't vary period
# like_syn.params['per3'].vary = False # Don't vary period
like_syn.params['dvdt'].vary = False # Don't vary dvdt
like_syn.params['curv'].vary = False # Don't vary curvature
