def test_hypermodel(dmx_psrs, caplog):
m2a = models.model_2a(dmx_psrs, noisedict=noise_dict)
m3a = models.model_3a(dmx_psrs, noisedict=noise_dict)
ptas = {0: m2a, 1: m3a}
hm = hypermodel.HyperModel(ptas)
assert hasattr(hm, 'get_lnlikelihood')
assert 'gw_log10_A' in hm.param_names
assert 'nmodel' in hm.param_names
def test_hyper_sampler(dmx_psrs, caplog):
m2a = models.model_2a(dmx_psrs, noisedict=noise_dict)
m3a = models.model_3a(dmx_psrs, noisedict=noise_dict)
ptas = {0: m2a, 1: m3a}
hm = hypermodel.HyperModel(ptas)
samp = hm.setup_sampler(outdir=outdir, human='tester')
assert hasattr(samp, "sample")
paramfile = os.path.join(outdir, "pars.txt")
assert os.path.isfile(paramfile)
with open(paramfile, "r") as f:
params = [line.rstrip('\n') for line in f]
for ptapar, filepar in zip(hm.param_names, params):
assert ptapar == filepar
x0 = hm.initial_sample()
assert len(x0) == len(hm.param_names)
std_models = [(s + rn_std + gw_std)(psr) for psr in pkl_psrs]
low_models = [(s + rn_low + gw_low)(psr) for psr in pkl_psrs]
pta_std = signal_base.PTA(std_models)
pta_low = signal_base.PTA(low_models)
ptas = {0:pta_std,
1:pta_low}
pta_std.set_default_params(noise)
pta_low.set_default_params(noise)
with open(args.pta_pkl,'wb') as fout:
cloudpickle.dump(ptas,fout)
hm = hypermodel.HyperModel(models=ptas)
sampler = hm.setup_sampler(outdir=args.outdir, resume=True,
empirical_distr = args.emp_distr)
try:
gw_std_freqs = get_freqs(pta_std, signal_id='gw')
gw_low_freqs = get_freqs(pta_low, signal_id='gw')
np.savetxt(args.outdir + 'low_gw_freqs.txt', gw_low_freqs, fmt='%.18e')
np.savetxt(args.outdir + 'std_gw_freqs.txt', gw_std_freqs, fmt='%.18e')
except:
pass
model_params = {}
for ii,mod in enumerate(ptas):
model_params.update({ii : ptas[ii].param_names})
mask = np.array([str(p) in param_diffs for p in model.params])
x0.extend([np.array(pp.sample()).ravel().tolist() \
if pp.name not in noisedict.keys() \
else np.array(noisedict[pp.name]).ravel().tolist() \
for pp in np.array(model.params)[mask]])
uniq_params = np.union1d([str(p) for p in model.params], \
uniq_params)
x0.extend([[0.1]])
return np.array([p for sublist in x0 for p in sublist])
if params.sampler == 'ptmcmcsampler':
super_model = hypermodel.HyperModel(pta)
print('Super model parameters: ', super_model.params)
print('Output directory: ', params.output_dir)
# Filling in PTMCMCSampler jump covariance matrix
# if params.mcmc_covm is not None:
# ndim = len(super_model.param_names)
# identity_covm = np.diag(np.ones(ndim) * 1**2)
# identity_covm_df = pd.DataFrame(identity_covm, \
# index=super_model.param_names, \
# columns=super_model.param_names)
# covm = params.mcmc_covm.combine_first(identity_covm_df)
# identity_covm_df.update(covm)
# covm = np.array(identity_covm_df)
# else:
# covm = None
sampler = super_model.setup_sampler(
n_gwbfreqs=args.nfreqs,
gamma_common=args.gamma_gw,
delta_common=0.,
upper_limit=False,
bayesephem=args.bayes_ephem,
be_type='setIII',
wideband=False,
rn_psrs=rn_psrs,
pshift=args.pshift)
ptas = {0: pta_crn, 1: pta_gw}
if args.model_wts is None:
model_wts = None
else:
model_wts = dict(enumerate(args.model_wts))
hm = hypermodel.HyperModel(models=ptas, log_weights=model_wts)
c0 = co.HyperModelCore(args.outdir, chaindir=args.outdir)
for ii in np.random.randint(c0.burn, c0.chain.shape[0], 100):
x = c0.chain[ii, :-4]
new_lnlike = hm.get_lnlikelihood(x)
new_lnpost = new_lnlike + hm.get_lnprior(x)
old_lnlike = c0.chain[ii, -3]
old_lnpost = c0.chain[ii, -4]
print('New LnLike, Old LnLike, New-Old:', new_lnlike, old_lnlike,
new_lnlike - old_lnlike)
print('New LnPost, Old LnPost, New-Old:', new_lnpost, old_lnpost,
new_lnpost - old_lnpost)
