def init_os_common(dataset, h5file, psrlist, nf, noisedir=None, noVaryNoise=False,
incJitterEquad=True, incEquad=True, outputfile=None):
model = PALmodels.PTAmodels(h5file, pulsars=psrlist)
# make model dictionary, can play with number of frequencies here
# Note: Tmax=0 means it will use the largest time span in the array to set the frequency bin size
fullmodel = model.makeModelDict(incRedNoise=True, incEquad=incEquad, incJitterEquad=incJitterEquad,
likfunc='mark9', incGWB=True, incEphemMixture=False, nfreqs=nf, Tmax=0)
# loop through pulsars and fix white noise parameters to values in file
if noisedir is not None:
for ct, p in enumerate(model.psr):
d = np.genfromtxt(noisedir + p.name + '_noise.txt', dtype='S42')
pars = d[:,0]
vals = np.array([float(d[ii,2]) for ii in range(d.shape[0])])
sigs = [psig for psig in fullmodel['signals'] if psig['pulsarind'] == ct]
sigs = PALutils.fixNoiseValues(sigs, vals, pars, bvary=False, verbose=False)
# turn back on red-noise parameters (will use later when drawing from posteriors)
for sig in fullmodel['signals']:
if sig['corr'] == 'single' and sig['stype'] == 'powerlaw':
sig['bvary'][1] = True
sig['bvary'][0] = True
if noVaryNoise:
nflags = ['efac', 'equad', 'jitter', 'jitter_equad']
for sig in fullmodel['signals']:
if sig['stype'] in nflags:
sig['bvary'][0] = False
# intialize mdoel
model.initModel(fullmodel, memsave=True, write='no')
# start parameters off at initial values (i.e the fixed values red in above)
p0 = model.initParameters(fixpstart=True)
# essentially turn off GWB component (again we will do something different when drawing from full PTA posterior)
p0[-2] = -19
# call once to fix white noise
xi, rho, sig, Opt, Sig = model.opt_stat_mark9(p0)
if outputfile is not None:
f = open(outputfile, 'w')
f.write('Opt {0}\n'.format(Opt))
f.write('Sig {0}\n'.format(Sig))
f.write('SNR {0}\n'.format(Opt/Sig))
f.close()
return model
def __init__(self,
chaindir,
h5file=None,
jsonfile=None,
outdir='noise_output',
save=True,
nreal=1000):
ret = get_mcmc_files(chaindir)
self.chain, self.pars, self.logp = ret[0], ret[1], ret[2]
if len(ret) == 4:
self.indicator = ret[3]
else:
self.indicator = None
self.ndim = len(self.pars)
if not os.path.exists(outdir):
try:
os.makedirs(outdir)
except OSError:
pass
self.jsonfile = jsonfile
self.h5file = h5file
if self.jsonfile is None:
self.model = None
else:
# open JSON file
with open(self.jsonfile) as f:
pmodel = OrderedDict(json.load(f))
pmodel['likfunc'] = 'mark6'
# load model
pulsars = list(map(str, pmodel['pulsarnames']))
self.model = PALmodels.PTAmodels(self.h5file, pulsars=pulsars)
self.model.initModel(pmodel, memsave=True, verbose=True)
p0 = self.model.initParameters()
self.model.mark6LogLikelihood(p0,
incCorrelations=False,
incJitter=True,
varyNoise=True)
self.outdir = outdir
self.save = save
if args.Tmax == 1:
args.Tmax = None
if not os.path.exists(args.outDir):
try:
os.makedirs(args.outDir)
except OSError:
pass
# open file
incGWB = args.incGWB
if args.pname[0] != 'all':
model = PALmodels.PTAmodels(args.h5file, pulsars=args.pname,
start_time=args.start_time,
end_time=args.end_time)
elif args.pname[0] == 'all':
print('Using all pulsars')
pulsars = 'all'
model = PALmodels.PTAmodels(args.h5file, pulsars=pulsars,
start_time=args.start_time,
end_time=args.end_time)
# model options
separateEfacs = args.separateEfacs == 'backend'
separateEfacsByFreq = args.separateEfacs == 'frequencies'
separateEquads = args.separateEquads == 'backend'
separateEquadsByFreq = args.separateEquads == 'frequencies'
C += np.ones((len(l), len(l))) * ecorr[l[0]]
avr = 1 / np.dot(M, np.dot(np.linalg.inv(C), M))
aveerr[i] = np.sqrt(avr)
averes[i] = avr * np.dot(M, np.dot(np.linalg.inv(C), res[l]))
if flags is not None:
return avetoas, averes, aveerr, aveflags
else:
return avetoas, aveerr, averes
h5file = './sns.h5'
pulsar = ['J1713+0747']
model = PALmodels.PTAmodels(h5file, pulsars=pulsar)
fullmodel = model.makeModelDict(incRedNoise=True,
noiseModel='powerlaw',
separateEfacsByFreq=True,
separateEquadsByFreq=True,
separateJitterEquadByFreq=True,
incEquad=True,
incJitterEquad=False,
likfunc='mark6')
for sig in fullmodel['signals']:
if sig['stype'] == 'efac':
#print sig['flagvalue']
#if np.any([e in sig['flagvalue'] for e in ['M4', 'M3', 'ABPP']]):
#if True:
sig['bvary'] = [False]
if not os.path.exists(chaindir):
try:
os.makedirs(chaindir)
except OSError:
pass
print "setting up run for PSR: " + this_PSR
#####
## Initialize Model that will include a search for:
## Basic Noise -- red noise (pow) + DM var (pow) + efac (default) + equad + ecorr (jitter)
## Band Noise -- + band noise (pow); need to determine band-binning
## System Noise -- + sys noise (pow); too many params, use subset of systems (most important only)
## NonStationary -- + wavelet/shapelet; non-stationary features in red/DM noise
## periodic DM -- + DM Amp and spec index as sinusoid
#####
model = PALmodels.PTAmodels(h5filename, pulsars=this_PSR)
basicModel = model.makeModelDict(incRedNoise=True,
noiseModel='powerlaw',
nfreqs=20,
incDM=True,
dmModel='powerlaw',
ndmfreqs=20,
incEquad=True,
incJitterEquad=jitter,
likfunc='mark9')
model.initModel(basicModel, memsave=True, fromFile=False,
verbose=True) # trying verbose=True to see output on cluster
params = model.get_varying_parameters() # names of params for triplot labels
print 'Run Params:'