Tspan=args.tspan)
x0 = np.hstack(p.sample() for p in pta.params)
ndim = x0.size
# initial jump covariance matrix
cov = np.diag(np.ones(ndim) * 0.01**2)
# set up jump groups by red noise groups
groups = model_utils.get_parameter_groups(pta)
sampler = ptmcmc(ndim,
pta.get_lnlikelihood,
pta.get_lnprior,
cov,
groups=groups,
outDir=args.outdir,
resume=True)
achrom_freqs = get_freqs(pta)
np.save(args.outdir + 'pars.npy', pta.param_names)
np.save(args.outdir + 'par_model.npy', np.array(pta.params).astype(str))
np.save(args.outdir + 'signals.npy', list(pta.signals.keys()))
np.savetxt(args.outdir + 'achrom_rn_freqs.txt', achrom_freqs, fmt='%.18e')
jp = model_utils.JumpProposal(pta)
sampler.addProposalToCycle(jp.draw_from_prior, 15)
sampler.addProposalToCycle(jp.draw_from_red_prior, 15)
sampler.addProposalToCycle(jp.draw_from_gwb_log_uniform_distribution, 15)
if args.bayes_ephem:
def setup_sampler(pta, outdir='chains', resume=False, empirical_distr=None):
"""
Sets up an instance of PTMCMC sampler.
We initialize the sampler the likelihood and prior function
from the PTA object. We set up an initial jump covariance matrix
with fairly small jumps as this will be adapted as the MCMC runs.
We will setup an output directory in `outdir` that will contain
the chain (first n columns are the samples for the n parameters
and last 4 are log-posterior, log-likelihood, acceptance rate, and
an indicator variable for parallel tempering but it doesn't matter
because we aren't using parallel tempering).
We then add several custom jump proposals to the mix based on
whether or not certain parameters are in the model. These are
all either draws from the prior distribution of parameters or
draws from uniform distributions.
"""
# dimension of parameter space
params = pta.param_names
ndim = len(params)
# initial jump covariance matrix
cov = np.diag(np.ones(ndim) * 0.1**2)
# parameter groupings
groups = get_parameter_groups(pta)
sampler = ptmcmc(ndim, pta.get_lnlikelihood, pta.get_lnprior, cov, groups=groups,
outDir=outdir, resume=resume)
np.savetxt(outdir+'/pars.txt',
list(map(str, pta.param_names)), fmt='%s')
np.savetxt(outdir+'/priors.txt',
list(map(lambda x: str(x.__repr__()), pta.params)), fmt='%s')
# additional jump proposals
jp = JumpProposal(pta, empirical_distr=empirical_distr)
# always add draw from prior
sampler.addProposalToCycle(jp.draw_from_prior, 5)
# try adding empirical proposals
if empirical_distr is not None:
print('Adding empirical proposals...\n')
sampler.addProposalToCycle(jp.draw_from_empirical_distr, 10)
# Red noise prior draw
if 'red noise' in jp.snames:
print('Adding red noise prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_red_prior, 10)
# DM GP noise prior draw
if 'dm_gp' in jp.snames:
print('Adding DM GP noise prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dm_gp_prior, 10)
# DM annual prior draw
if 'dm_s1yr' in jp.snames:
print('Adding DM annual prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dm1yr_prior, 10)
# DM dip prior draw
if 'dmexp' in jp.snames:
print('Adding DM exponential dip prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmexpdip_prior, 10)
# DM cusp prior draw
if 'dm_cusp' in jp.snames:
print('Adding DM exponential cusp prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmexpcusp_prior, 10)
# DMX prior draw
if 'dmx_signal' in jp.snames:
print('Adding DMX prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmx_prior, 10)
# Ephemeris prior draw
if 'd_jupiter_mass' in pta.param_names:
print('Adding ephemeris model prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_ephem_prior, 10)
# GWB uniform distribution draw
if 'gw_log10_A' in pta.param_names:
print('Adding GWB uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_gwb_log_uniform_distribution, 10)
# Dipole uniform distribution draw
if 'dipole_log10_A' in pta.param_names:
print('Adding dipole uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_dipole_log_uniform_distribution, 10)
# Monopole uniform distribution draw
if 'monopole_log10_A' in pta.param_names:
print('Adding monopole uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_monopole_log_uniform_distribution, 10)
# Altpol uniform distribution draw
if 'log10Apol_tt' in pta.param_names:
print('Adding alternative GW-polarization uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_altpol_log_uniform_distribution, 10)
# BWM prior draw
if 'bwm_log10_A' in pta.param_names:
print('Adding BWM prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_bwm_prior, 10)
# CW prior draw
if 'cw_log10_h' in pta.param_names:
print('Adding CW strain prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_cw_log_uniform_distribution, 10)
if 'cw_log10_Mc' in pta.param_names:
print('Adding CW prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_cw_distribution, 10)
return sampler
def setup_sampler(self, outdir='chains', resume=False, sample_nmodel=True,
empirical_distr=None, groups=None):
"""
Sets up an instance of PTMCMC sampler.
We initialize the sampler the likelihood and prior function
from the PTA object. We set up an initial jump covariance matrix
with fairly small jumps as this will be adapted as the MCMC runs.
We will setup an output directory in `outdir` that will contain
the chain (first n columns are the samples for the n parameters
and last 4 are log-posterior, log-likelihood, acceptance rate, and
an indicator variable for parallel tempering but it doesn't matter
because we aren't using parallel tempering).
We then add several custom jump proposals to the mix based on
whether or not certain parameters are in the model. These are
all either draws from the prior distribution of parameters or
draws from uniform distributions.
"""
# dimension of parameter space
ndim = len(self.param_names)
# initial jump covariance matrix
cov = np.diag(np.ones(ndim) * 1**2) ## used to be 0.1
# parameter groupings
if groups is None:
groups = self.get_parameter_groups()
sampler = ptmcmc(ndim, self.get_lnlikelihood, self.get_lnprior, cov,
groups=groups, outDir=outdir, resume=resume)
np.savetxt(outdir+'/pars.txt', self.param_names, fmt='%s')
np.savetxt(outdir+'/priors.txt', self.params, fmt='%s')
# additional jump proposals
jp = JumpProposal(self, self.snames, empirical_distr=empirical_distr)
# always add draw from prior
sampler.addProposalToCycle(jp.draw_from_prior, 5)
# try adding empirical proposals
if empirical_distr is not None:
print('Adding empirical proposals...\n')
sampler.addProposalToCycle(jp.draw_from_empirical_distr, 25)
# Red noise prior draw
if 'red noise' in self.snames:
print('Adding red noise prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_red_prior, 10)
# DM GP noise prior draw
if 'dm_gp' in self.snames:
print('Adding DM GP noise prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dm_gp_prior, 10)
# DM annual prior draw
if 'dm_s1yr' in jp.snames:
print('Adding DM annual prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dm1yr_prior, 10)
# DM dip prior draw
if 'dmexp' in '\t'.join(jp.snames):
print('Adding DM exponential dip prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmexpdip_prior, 10)
# DM cusp prior draw
if 'dm_cusp' in jp.snames:
print('Adding DM exponential cusp prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmexpcusp_prior, 10)
# DMX prior draw
if 'dmx_signal' in jp.snames:
print('Adding DMX prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmx_prior, 10)
# SW prior draw
if 'gp_sw' in jp.snames:
print('Adding Solar Wind DM GP prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dm_sw_prior, 10)
# Ephemeris prior draw
if 'd_jupiter_mass' in self.param_names:
print('Adding ephemeris model prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_ephem_prior, 10)
# GWB uniform distribution draw
if 'gw_log10_A' in self.param_names:
print('Adding GWB uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_gwb_log_uniform_distribution, 10)
# Dipole uniform distribution draw
if 'dipole_log10_A' in self.param_names:
print('Adding dipole uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_dipole_log_uniform_distribution, 10)
# Monopole uniform distribution draw
if 'monopole_log10_A' in self.param_names:
print('Adding monopole uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_monopole_log_uniform_distribution, 10)
# BWM prior draw
if 'bwm_log10_A' in self.param_names:
print('Adding BWM prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_bwm_prior, 10)
# FDM prior draw
if 'fdm_log10_A' in self.param_names:
print('Adding FDM prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_fdm_prior, 10)
# CW prior draw
if 'cw_log10_h' in self.param_names:
print('Adding CW prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_cw_log_uniform_distribution, 10)
# Prior distribution draw for parameters named GW
if any([str(p).split(':')[0] for p in list(self.params) if 'gw' in str(p)]):
print('Adding gw param prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_par_prior(
par_names=[str(p).split(':')[0] for
p in list(self.params)
if 'gw' in str(p)]), 10)
# Model index distribution draw
if sample_nmodel:
if 'nmodel' in self.param_names:
print('Adding nmodel uniform distribution draws...\n')
sampler.addProposalToCycle(self.draw_from_nmodel_prior, 25)
return sampler
def setup_sampler(pta, outdir='chains', resume=False):
"""
Sets up an instance of PTMCMC sampler.
We initialize the sampler the likelihood and prior function
from the PTA object. We set up an initial jump covariance matrix
with fairly small jumps as this will be adapted as the MCMC runs.
We will setup an output directory in `outdir` that will contain
the chain (first n columns are the samples for the n parameters
and last 4 are log-posterior, log-likelihood, acceptance rate, and
an indicator variable for parallel tempering but it doesn't matter
because we aren't using parallel tempering).
We then add several custom jump proposals to the mix based on
whether or not certain parameters are in the model. These are
all either draws from the prior distribution of parameters or
draws from uniform distributions.
"""
# dimension of parameter space
ndim = len(pta.param_names)
# initial jump covariance matrix
cov = np.diag(np.ones(ndim) * 0.1**2)
# parameter groupings
groups = get_parameter_groups(pta)
sampler = ptmcmc(ndim,
pta.get_lnlikelihood,
pta.get_lnprior,
cov,
groups=groups,
outDir=outdir,
resume=resume)
np.savetxt(outdir + '/pars.txt', list(map(str, pta.param_names)), fmt='%s')
np.savetxt(outdir + '/priors.txt',
list(map(lambda x: str(x.__repr__()), pta.params)),
fmt='%s')
# additional jump proposals
jp = JumpProposal(pta)
# always add draw from prior
sampler.addProposalToCycle(jp.draw_from_prior, 5)
# Ephemeris prior draw
if 'd_jupiter_mass' in pta.param_names:
print('Adding ephemeris model prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_ephem_prior, 10)
# GWB uniform distribution draw
if 'log10_A_gw' in pta.param_names:
print('Adding GWB uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_gwb_log_uniform_distribution,
10)
# Dipole uniform distribution draw
if 'log10_A_dipole' in pta.param_names:
print('Adding dipole uniform distribution draws...\n')
sampler.addProposalToCycle(
jp.draw_from_dipole_log_uniform_distribution, 10)
# GWB uniform distribution draw
if 'log10_A_monopole' in pta.param_names:
print('Adding monopole uniform distribution draws...\n')
sampler.addProposalToCycle(
jp.draw_from_monopole_log_uniform_distribution, 10)
return sampler
# initial jump covariance matrix
cov = np.diag(np.ones(ndim) * 0.01**2)
# Now we figure out which indices the red noise parameters have
rn_idx1 = pta.param_names.index(psr.name + '_red_noise_log10_A')
rn_idx2 = pta.param_names.index(psr.name + '_red_noise_gamma')
# set up jump groups by red noise groups
ndim = len(xs)
groups = [range(0, ndim)]
groups.extend([[rn_idx1, rn_idx2]])
# intialize sampler
sampler = ptmcmc(ndim,
pta.get_lnlikelihood,
pta.get_lnprior,
cov,
groups=groups,
outDir=psroutdir)
# sampler for N steps
N = int(1e6)
x0 = np.hstack(p.sample() for p in pta.params)
sampler.sample(x0, N, SCAMweight=30, AMweight=15, DEweight=50)
chain = np.loadtxt(psroutdir + 'chain_1.txt')
pars = sorted(xs.keys())
burn = int(0.25 * chain.shape[0])
#make dictionary of pulsar parameters from these runs
psr_dict[psr.name] = {}
for idx, param_name in enumerate(pars):
def setup_sampler(pta,
outdir='chains',
resume=False,
empirical_distr=None,
groups=None,
human=None):
"""
Sets up an instance of PTMCMC sampler.
We initialize the sampler the likelihood and prior function
from the PTA object. We set up an initial jump covariance matrix
with fairly small jumps as this will be adapted as the MCMC runs.
We will setup an output directory in `outdir` that will contain
the chain (first n columns are the samples for the n parameters
and last 4 are log-posterior, log-likelihood, acceptance rate, and
an indicator variable for parallel tempering but it doesn't matter
because we aren't using parallel tempering).
We then add several custom jump proposals to the mix based on
whether or not certain parameters are in the model. These are
all either draws from the prior distribution of parameters or
draws from uniform distributions.
"""
# dimension of parameter space
params = pta.param_names
ndim = len(params)
# initial jump covariance matrix
if os.path.exists(outdir + '/cov.npy'):
cov = np.load(outdir + '/cov.npy')
else:
cov = np.diag(np.ones(ndim) * 0.1**2)
# parameter groupings
if groups is None:
groups = get_parameter_groups(pta)
sampler = ptmcmc(ndim,
pta.get_lnlikelihood,
pta.get_lnprior,
cov,
groups=groups,
outDir=outdir,
resume=resume)
save_runtime_info(pta, sampler.outDir, human)
# additional jump proposals
jp = JumpProposal(pta, empirical_distr=empirical_distr)
sampler.jp = jp
# always add draw from prior
sampler.addProposalToCycle(jp.draw_from_prior, 5)
# try adding empirical proposals
if empirical_distr is not None:
print('Attempting to add empirical proposals...\n')
sampler.addProposalToCycle(jp.draw_from_empirical_distr, 10)
# Red noise prior draw
if 'red noise' in jp.snames:
print('Adding red noise prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_red_prior, 10)
# DM GP noise prior draw
if 'dm_gp' in jp.snames:
print('Adding DM GP noise prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dm_gp_prior, 10)
# DM annual prior draw
if 'dm_s1yr' in jp.snames:
print('Adding DM annual prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dm1yr_prior, 10)
# DM dip prior draw
if 'dmexp' in jp.snames:
print('Adding DM exponential dip prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmexpdip_prior, 10)
# DM cusp prior draw
if 'dm_cusp' in jp.snames:
print('Adding DM exponential cusp prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmexpcusp_prior, 10)
# DMX prior draw
if 'dmx_signal' in jp.snames:
print('Adding DMX prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_dmx_prior, 10)
# Ephemeris prior draw
if 'd_jupiter_mass' in pta.param_names:
print('Adding ephemeris model prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_ephem_prior, 10)
# GWB uniform distribution draw
if np.any([('gw' in par and 'log10_A' in par) for par in pta.param_names]):
print('Adding GWB uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_gwb_log_uniform_distribution,
10)
# Dipole uniform distribution draw
if 'dipole_log10_A' in pta.param_names:
print('Adding dipole uniform distribution draws...\n')
sampler.addProposalToCycle(
jp.draw_from_dipole_log_uniform_distribution, 10)
# Monopole uniform distribution draw
if 'monopole_log10_A' in pta.param_names:
print('Adding monopole uniform distribution draws...\n')
sampler.addProposalToCycle(
jp.draw_from_monopole_log_uniform_distribution, 10)
# Altpol uniform distribution draw
if 'log10Apol_tt' in pta.param_names:
print(
'Adding alternative GW-polarization uniform distribution draws...\n'
)
sampler.addProposalToCycle(
jp.draw_from_altpol_log_uniform_distribution, 10)
# print('Adding alternative GW-polarization uniform distribution draws...\n')
# sampler.addProposalToCycle(jp.draw_from_polar_log_uniform_distribution, 10)
# TT uniform distribution draw
if 'log10_A_TT' in pta.param_names:
print('Adding TT uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_TT_log_uniform_distribution,
10)
# ST uniform distribution draw
if 'log10_A_ST' in pta.param_names:
print('Adding ST uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_ST_log_uniform_distribution,
10)
# VL uniform distribution draw
if 'log10_A_VL' in pta.param_names:
print('Adding VL uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_VL_log_uniform_distribution,
10)
# SL uniform distribution draw
if 'log10_A_SL' in pta.param_names:
print('Adding SL uniform distribution draws...\n')
sampler.addProposalToCycle(jp.draw_from_SL_log_uniform_distribution,
10)
# BWM prior draw
if 'bwm_log10_A' in pta.param_names:
print('Adding BWM prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_bwm_prior, 10)
# FDM prior draw
if 'fdm_log10_A' in pta.param_names:
print('Adding FDM prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_fdm_prior, 10)
# CW prior draw
if 'cw_log10_h' in pta.param_names:
print('Adding CW strain prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_cw_log_uniform_distribution,
10)
if 'cw_log10_Mc' in pta.param_names:
print('Adding CW prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_cw_distribution, 10)
# SL uniform distribution draw
if 'bin_orf_A1' in pta.param_names:
print('Adding bin-orf prior draws...\n')
sampler.addProposalToCycle(jp.draw_from_bin_orf_uniform_distribution,
10)
return sampler
## sampler ##
###############
# dimension of parameter space
x0 = np.hstack(p.sample() for p in pta.params)
ndim = len(x0)
# initial jump covariance matrix
cov = np.diag(np.ones(ndim) * 0.1**2)
# parameter groupings
groups = get_parameter_groups(pta)
sampler = ptmcmc(ndim,
pta.get_lnlikelihood,
pta.get_lnprior,
cov,
groups=groups,
outDir=rundir,
resume=False)
# add prior draws to proposal cycle
jp = JumpProposal(pta)
sampler.addProposalToCycle(jp.draw_from_prior, 5)
sampler.addProposalToCycle(jp.draw_from_bwm_prior, 5)
if amp_prior == 'uniform':
draw_bwm_loguni = jp.build_log_uni_draw('bwm_log10_A',
logminA,
logmaxA,
logparam=True)
sampler.addProposalToCycle(draw_bwm_loguni, 10)
t0=par[9],
epsilon=par[10],
zeta=par[11],
eta=par[12],
Ms=par[13],
theta=par[14],
sigma=par[15],
e0=par[16],
rho=par[17])
model = mr.mergerrate(M1, q, z, f, **initpar).hmodelt(fbin=None)[0]
l = -0.5 / (data[:, 5] * data[:, 5]) * (model - data[:, 2]) * (
model - data[:, 2]) - 0.5 * np.log10(2 * np.pi) #detection
return np.sum(l)
x0 = np.array([
-2.8, -0.2, 11.25, -1.25, 0., 0.025, 0.8, 0., 0., 1., 0., -0.5, 0., 8.25,
1., 0.4, 0.5, 0.
])
ndim = len(x0)
cov = np.diag(np.ones(ndim) * 0.01**2)
N = 1000000
sampler = ptmcmc(ndim,
log_likelihood,
log_prior,
cov,
outDir='./output',
resume=False)
sampler.sample(x0, N, SCAMweight=30, AMweight=15, DEweight=50)