def plots(args):
"""
Generate plots
Args:
args (ArgumentParser): command line arguments
"""
config_file = args.setupfn
conf_base = os.path.basename(config_file).split('.')[0]
statfile = os.path.join(
args.outputdir, "{}_radvel.stat".format(conf_base)
)
status = load_status(statfile)
P, post = radvel.utils.initialize_posterior(config_file,
decorr=args.decorr)
assert status.getboolean('fit', 'run'), \
"Must perform max-liklihood fit before plotting"
post = radvel.posterior.load(status.get('fit', 'postfile'))
for ptype in args.type:
print("Creating {} plot for {}".format(ptype, conf_base))
if ptype == 'rv':
args.plotkw['uparams'] = post.uparams
args.plotkw['status'] = status
if 'saveplot' not in args.plotkw:
saveto = os.path.join(
args.outputdir, conf_base+'_rv_multipanel.pdf'
)
else:
saveto = args.plotkw['saveplot']
args.plotkw.pop('saveplot')
P, _ = radvel.utils.initialize_posterior(config_file)
if hasattr(P, 'bjd0'):
args.plotkw['epoch'] = P.bjd0
if args.gp:
GPPlot = orbit_plots.GPMultipanelPlot(
post, saveplot=saveto, **args.plotkw
)
GPPlot.plot_multipanel()
else:
RVPlot = orbit_plots.MultipanelPlot(
post, saveplot=saveto, **args.plotkw
)
RVPlot.plot_multipanel()
# check to make sure that Posterior is not GP, print warning if it is
if isinstance(post.likelihood, radvel.likelihood.CompositeLikelihood):
like_list = post.likelihood.like_list
else:
like_list = [post.likelihood]
for like in like_list:
if isinstance(like, radvel.likelihood.GPLikelihood):
print("WARNING: GP Likelihood(s) detected. \
You may want to use the '--gp' flag when making these plots.")
break
if ptype == 'corner' or ptype == 'auto' or ptype == 'trend':
assert status.getboolean('mcmc', 'run'), \
"Must run MCMC before making corner, auto, or trend plots"
chains = pd.read_csv(status.get('mcmc', 'chainfile'))
autocorr = pd.read_csv(status.get('mcmc', 'autocorrfile'))
if ptype == 'auto':
saveto = os.path.join(args.outputdir, conf_base+'_auto.pdf')
Auto = mcmc_plots.AutoPlot(autocorr, saveplot=saveto)
Auto.plot()
if ptype == 'corner':
saveto = os.path.join(args.outputdir, conf_base+'_corner.pdf')
Corner = mcmc_plots.CornerPlot(post, chains, saveplot=saveto)
Corner.plot()
if ptype == 'trend':
nwalkers = status.getint('mcmc', 'nwalkers')
nensembles = status.getint('mcmc', 'nensembles')
saveto = os.path.join(args.outputdir, conf_base+'_trends.pdf')
Trend = mcmc_plots.TrendPlot(post, chains, nwalkers, nensembles, saveto)
Trend.plot()
if ptype == 'derived':
assert status.has_section('derive'), \
"Must run `radvel derive` before plotting derived parameters"
P, _ = radvel.utils.initialize_posterior(config_file)
chains = pd.read_csv(status.get('derive', 'chainfile'))
saveto = os.path.join(
args.outputdir, conf_base+'_corner_derived_pars.pdf'
)
Derived = mcmc_plots.DerivedPlot(chains, P, saveplot=saveto)
Derived.plot()
savestate = {'{}_plot'.format(ptype): os.path.relpath(saveto)}
save_status(statfile, 'plot', savestate)
def run_search(self,
fixed_threshold=None,
outdir=None,
mkoutdir=True,
running=True):
"""Run an iterative search for planets not given in posterior.
Args:
fixed_threshold (float): (optional) use a fixed delta BIC threshold
mkoutdir (bool): create the output directory?
"""
if outdir is None:
outdir = os.path.join(os.getcwd(), self.starname)
if mkoutdir and not os.path.exists(outdir):
os.mkdir(outdir)
if self.trend:
self.trend_test()
run = True
while run:
if self.num_planets != 0:
self.add_planet()
perioder = periodogram.Periodogram(self.post,
basebic=self.basebic,
minsearchp=self.min_per,
maxsearchp=self.max_per,
fap=self.fap,
manual_grid=self.manual_grid,
oversampling=self.oversampling,
baseline=self.baseline,
eccentric=self.eccentric,
workers=self.workers,
verbose=self.verbose)
# Run the periodogram, store arrays and threshold (if computed).
perioder.per_bic()
self.periodograms[self.num_planets] = perioder.power[self.crit]
if self.num_planets == 0 or self.pers is None:
self.pers = perioder.pers
if fixed_threshold is None:
perioder.eFAP()
self.eFAPs[self.num_planets] = perioder.fap_min
else:
perioder.bic_thresh = fixed_threshold
self.bic_threshes[self.num_planets] = perioder.bic_thresh
self.best_bics[self.num_planets] = perioder.best_bic
if self.save_outputs:
perioder.plot_per()
perioder.fig.savefig(outdir +
'/dbic{}.pdf'.format(self.num_planets +
1))
# Check whether there is a detection. If so, fit free and proceed.
if perioder.best_bic > perioder.bic_thresh:
self.num_planets += 1
for k in self.post.params.keys():
self.post.params[k].value = perioder.bestfit_params[k]
# Generalize tc reset to each new discovery.
tckey = 'tc{}'.format(self.num_planets)
if self.post.params[tckey].value < np.amin(self.data.time):
self.post.params[tckey].value = np.median(self.data.time)
for n in np.arange(1, self.num_planets + 1):
self.post.params['k{}'.format(n)].vary = False
self.post.params['per{}'.format(n)].vary = False
self.post.params['secosw{}'.format(n)].vary = False
self.post.params['secosw{}'.format(n)].vary = False
if n != self.num_planets:
self.post.params['tc{}'.format(n)].vary = False
self.post = radvel.fitting.maxlike_fitting(self.post,
verbose=False)
for n in np.arange(1, self.num_planets + 1):
self.post.params['k{}'.format(n)].vary = True
self.post.params['per{}'.format(n)].vary = True
self.post.params['secosw{}'.format(n)].vary = True
self.post.params['secosw{}'.format(n)].vary = True
self.post.params['tc{}'.format(n)].vary = True
self.fit_orbit()
self.all_params.append(self.post.params)
self.basebic = self.post.likelihood.bic()
else:
self.sub_planet()
# 8/3: Update the basebic anyway, for injections.
self.basebic = self.post.likelihood.bic()
run = False
if self.num_planets >= self.max_planets:
run = False
# If any jitter values are negative, flip them.
for key in self.post.params.keys():
if 'jit' in key:
if self.post.params[key].value < 0:
self.post.params[
key].value = -self.post.params[key].value
# Generate an orbit plot.
if self.save_outputs:
rvplot = orbit_plots.MultipanelPlot(
self.post,
saveplot=outdir +
'/orbit_plot{}.pdf'.format(self.num_planets))
multiplot_fig, ax_list = rvplot.plot_multipanel()
multiplot_fig.savefig(
outdir + '/orbit_plot{}.pdf'.format(self.num_planets))
# Generate running periodograms.
if running:
self.running_per()
# Run MCMC on final posterior, save new parameters and uncertainties.
if self.mcmc == True and (self.num_planets != 0
or self.post.params['dvdt'].vary == True):
self.post.uparams = {}
self.post.medparams = {}
self.post.maxparams = {}
# Use recommended parameters for mcmc.
nensembles = np.min([self.workers, 16])
if os.cpu_count() < nensembles:
nensembles = os.cpu_count()
# Set custom mcmc scales for e/w parameters.
for n in np.arange(1, self.num_planets + 1):
self.post.params['secosw{}'.format(n)].mcmcscale = 0.005
self.post.params['sesinw{}'.format(n)].mcmcscale = 0.005
# Sample in log-period space.
logpost = copy.deepcopy(self.post)
logparams = logpost.params.basis.to_any_basis(
logpost.params, 'logper tc secosw sesinw k')
logpost = utils.initialize_post(self.data, params=logparams)
# Run MCMC. #self.post #logpost
chains = radvel.mcmc(logpost,
nwalkers=50,
nrun=25000,
burnGR=1.03,
maxGR=1.0075,
minTz=2000,
minAfactor=15,
maxArchange=0.07,
burnAfactor=15,
minsteps=12500,
minpercent=50,
thin=5,
save=False,
ensembles=nensembles)
# Convert chains to per, e, w basis.
synthchains = logpost.params.basis.to_synth(chains)
synthquants = synthchains.quantile([0.159, 0.5, 0.841])
logpost = None
# Compress, thin, and save chains, in fitting and synthetic bases.
csvfn = outdir + '/chains.csv.tar.bz2'
synthchains.to_csv(csvfn, compression='bz2')
# Retrieve e and w medians & uncertainties from synthetic chains.
for n in np.arange(1, self.num_planets + 1):
e_key = 'e{}'.format(n)
w_key = 'w{}'.format(n)
# Add period if it's a synthetic parameter.
per_key = 'per{}'.format(n)
logper_key = 'logper{}'.format(n)
med_e = synthquants[e_key][0.5]
high_e = synthquants[e_key][0.841] - med_e
low_e = med_e - synthquants[e_key][0.159]
err_e = np.mean([high_e, low_e])
err_e = radvel.utils.round_sig(err_e)
med_e, err_e, errhigh_e = radvel.utils.sigfig(med_e, err_e)
max_e, err_e, errhigh_e = radvel.utils.sigfig(
self.post.params[e_key].value, err_e)
med_w = synthquants[w_key][0.5]
high_w = synthquants[w_key][0.841] - med_w
low_w = med_w - synthquants[w_key][0.159]
err_w = np.mean([high_w, low_w])
err_w = radvel.utils.round_sig(err_w)
med_w, err_w, errhigh_w = radvel.utils.sigfig(med_w, err_w)
max_w, err_w, errhigh_w = radvel.utils.sigfig(
self.post.params[w_key].value, err_w)
self.post.uparams[e_key] = err_e
self.post.uparams[w_key] = err_w
self.post.medparams[e_key] = med_e
self.post.medparams[w_key] = med_w
self.post.maxparams[e_key] = max_e
self.post.maxparams[w_key] = max_w
# Retrieve medians & uncertainties for the fitting basis parameters.
for par in self.post.params.keys():
if self.post.params[par].vary:
med = synthquants[par][0.5]
high = synthquants[par][0.841] - med
low = med - synthquants[par][0.159]
err = np.mean([high, low])
err = radvel.utils.round_sig(err)
med, err, errhigh = radvel.utils.sigfig(med, err)
max, err, errhigh = radvel.utils.sigfig(
self.post.params[par].value, err)
self.post.uparams[par] = err
self.post.medparams[par] = med
self.post.maxparams[par] = max
# Add uncertainties on derived parameters, if mass is provided.
if self.mstar is not None:
self.post = utils.derive(self.post, synthchains, self.mstar,
self.mstar_err)
if self.save_outputs:
# Generate a corner plot, sans nuisance parameters.
labels = []
for n in np.arange(1, self.num_planets + 1):
labels.append('per{}'.format(n))
labels.append('tc{}'.format(n))
labels.append('k{}'.format(n))
labels.append('secosw{}'.format(n))
labels.append('sesinw{}'.format(n))
if self.post.params['dvdt'].vary == True:
labels.append('dvdt')
if self.post.params['curv'].vary == True:
labels.append('curv')
texlabels = [
self.post.params.tex_labels().get(l, l) for l in labels
]
plot = corner.corner(synthchains[labels],
labels=texlabels,
label_kwargs={"fontsize": 14},
plot_datapoints=False,
bins=30,
quantiles=[0.16, 0.5, 0.84],
title_kwargs={"fontsize": 14},
show_titles=True,
smooth=True)
pl.savefig(outdir +
'/{}_corner_plot.pdf'.format(self.starname))
# Generate an orbit plot wth median parameters and uncertainties.
rvplot = orbit_plots.MultipanelPlot(
self.post,
saveplot=outdir +
'/orbit_plot_mc_{}.pdf'.format(self.starname),
uparams=self.post.uparams)
multiplot_fig, ax_list = rvplot.plot_multipanel()
multiplot_fig.savefig(
outdir + '/orbit_plot_mc_{}.pdf'.format(self.starname))
if self.save_outputs:
self.save(filename=outdir + '/post_final.pkl')
pickle_out = open(outdir + '/search.pkl', 'wb')
pickle.dump(self, pickle_out)
pickle_out.close()
periodograms_plus_pers = np.append([self.pers],
list(
self.periodograms.values()),
axis=0).T
np.savetxt(outdir + '/pers_periodograms.csv',
periodograms_plus_pers,
header='period BIC_array')
threshs_bics_faps = np.append(
[list(self.bic_threshes.values())],
[list(self.best_bics.values()),
list(self.eFAPs.values())],
axis=0).T
np.savetxt(outdir + '/thresholds_bics_faps.csv',
threshs_bics_faps,
header='threshold best_bic fap')
def _hack_radvel_plots(args):
"""
Hacked version of radvel.driver.plots
Args:
args (ArgumentParser): command line arguments
"""
config_file = args.setupfn
conf_base = os.path.basename(config_file).split('.')[0]
statfile = os.path.join(args.inputdir, "{}_radvel.stat".format(conf_base))
status = load_status(statfile)
assert status.getboolean('fit', 'run'), \
"Must perform max-liklihood fit before plotting"
post = radvel.posterior.load(status.get('fit', 'postfile'))
# update the plotted posterior to match the median best-fit parameters.
summarycsv = "../results/rv_fitting/t10onlytwo_LGB_20190228_fix_gammaddot/WASP4_post_summary.csv"
sdf = pd.read_csv(summarycsv)
for param in [c for c in sdf.columns if 'Unnamed' not in c]:
post.params[param] = radvel.Parameter(value=sdf.ix[1][param])
for ptype in args.type:
print("Creating {} plot for {}".format(ptype, conf_base))
if ptype == 'rv':
args.plotkw['uparams'] = post.uparams
saveto = os.path.join(args.outputdir,
conf_base + '_rv_multipanel.pdf')
P, _ = radvel.utils.initialize_posterior(config_file)
if hasattr(P, 'bjd0'):
args.plotkw['epoch'] = P.bjd0
#FIXME FIXME: see line 103 of this for how to access residuals
model = post.likelihood.model
rvtimes = post.likelihood.x
rverr = post.likelihood.errorbars()
num_planets = model.num_planets
#FIXME FIXME
rawresid = post.likelihood.residuals()
resid = (rawresid + post.params['dvdt'].value *
(rvtimes - model.time_base) + post.params['curv'].value *
(rvtimes - model.time_base)**2)
import IPython
IPython.embed()
RVPlot = orbit_plots.MultipanelPlot(post,
saveplot=saveto,
**args.plotkw)
RVPlot.plot_multipanel()
# check to make sure that Posterior is not GP, print warning if it is
if isinstance(post.likelihood,
radvel.likelihood.CompositeLikelihood):
like_list = post.likelihood.like_list
else:
like_list = [post.likelihood]
for like in like_list:
if isinstance(like, radvel.likelihood.GPLikelihood):
print("WARNING: GP Likelihood(s) detected."
"You may want to use the '--gp' flag"
"when making these plots.")
break
else:
raise NotImplementedError('nope')
savestate = {'{}_plot'.format(ptype): os.path.relpath(saveto)}
save_status(statfile, 'plot', savestate)
]
# post.priors += [radvel.prior.EccentricityPrior(3)]
print("Before fitting\n")
print(post, "\n")
# res = optimize.minimize(post.neglogprob_array, post.get_vary_params(), method='Nelder-Mead' )
res = optimize.minimize(post.neglogprob_array,
post.get_vary_params(),
method='L-BFGS-B')
# sf.plot_radvel_results(post.likelihood, t)
plt.savefig(fig_name + "%d_afterpriors.png" % err_ind, bbox_inches="tight")
print("After fitting\n")
print(post, "\n")
RVPlot = orbit_plots.MultipanelPlot(post)
RVPlot.plot_multipanel()
# plt.show()
plt.savefig(fig_name + "%d_afterpriors_multi.png" % err_ind,
bbox_inches="tight")
# plt.close(fig="all")
# df3 = radvel.mcmc(post) # amount of steps = nrun * 8 * nwalkers ?, default nrun is 10000, default nwalkers is 50?
df3 = radvel.mcmc(post, nrun=1000)
df3.to_pickle(fig_name + "%d_mcmc.pkl" % err_ind)
fig = corner.corner(df3[labels],
labels=labels,
quantiles=[0.16, 0.84],
plot_datapoints=False)
plt.savefig(fig_name + "%d_corner.png" % err_ind, bbox_inches="tight")