def test_emcee(self):
self.initialize(True)
nwalkers = 20
nsteps = 20
retriever = Retriever()
result = retriever.run_emcee(self.wavelength_bins,
self.depths,
self.errors,
self.fit_info,
nsteps=nsteps,
nwalkers=nwalkers,
include_condensation=False)
self.assertTrue(isinstance(result, emcee.ensemble.EnsembleSampler))
self.assertTrue(result.chain.shape,
(nwalkers, nsteps, len(self.fit_info.fit_param_names)))
self.assertTrue(result.lnprobability.shape, (nwalkers, nsteps))
retriever = Retriever()
result = retriever.run_emcee(self.wavelength_bins,
self.depths,
self.errors,
self.fit_info,
nsteps=nsteps,
nwalkers=nwalkers,
include_condensation=True,
plot_best=True)
self.assertTrue(isinstance(result, emcee.ensemble.EnsembleSampler))
self.assertEqual(
result.chain.shape,
(nwalkers, nsteps, len(self.fit_info.fit_param_names)))
self.assertEqual(result.lnprobability.shape, (nwalkers, nsteps))
fit_info.add_uniform_fit_param("log_cloudtop_P", -0.99, 5)
fit_info.add_uniform_fit_param("error_multiple", 0.5, 5)
#Use Nested Sampling to do the fitting
# with ThreadPoolExecutor() as executor:
# with ProcessPoolExecutor() as executor:
time_stamp = datetime.utcnow().strftime("%Y%m%d%H%M%S")
if bayesian_model == 'multinest':
with Pool(cpu_count()) as executor:
result = retriever.run_multinest(wave_bins, depths, errors, fit_info, nestle_kwargs={'pool':executor})#, 'bootstrap':0 # bootstrap for `dynesty`
result_dict = {'samples':result.samples, 'weights':result.weights, 'logl':result.logl}
joblib.dump(result_dict, 'multinest_results_{}.joblib.save'.format(time_stamp))
elif bayesian_model == 'emcee':
result = retriever.run_emcee(wave_bins, depths, errors, fit_info, nwalkers=nwalkers, nsteps=nsteps)
result_dict = {'flatchain':result.flatchain, 'flatlnprob':result.flatlnprobability, 'chain':result.chain, 'lnprob':result.lnprobability}
joblib.dump(result_dict, 'emcee_results_{}walkers_{}steps_{}.joblib.save'.format(nwalkers, nsteps, time_stamp))
else:
raise ValueError("Options for `bayesian_model` (-bm, --bayesianmodel) must be either 'multinest' or 'emcee'")
# Establish the Range in Wavelength to plot high resolution figures
wave_min = wave_bins.min()
wave_max = wave_bins.max()
n_theory_pts = 500
wavelengths_theory = np.linspace(wave_min, wave_max, n_theory_pts)
half_diff_lam = 0.5*np.median(np.diff(wavelengths_theory))
# Setup calculator to use the theoretical wavelengths
log_scatt_factor=0, scatt_slope=4, error_multiple=1, T_star=6091)
#Add fitting parameters - this specifies which parameters you want to fit
#e.g. since we have not included cloudtop_P, it will be fixed at the value specified in the constructor
fit_info.add_gaussian_fit_param('Rs', 0.02*R_sun)
fit_info.add_gaussian_fit_param('Mp', 0.04*M_jup)
# Here, emcee is initialized with walkers where R is between 0.9*R_guess and
# 1.1*R_guess. However, the hard limit on R is from 0 to infinity.
fit_info.add_uniform_fit_param('R', 0, np.inf, 0.9*R_guess, 1.1*R_guess)
fit_info.add_uniform_fit_param('T', 300, 3000, 0.5*T_guess, 1.5*T_guess)
fit_info.add_uniform_fit_param("log_scatt_factor", 0, 5, 0, 1)
fit_info.add_uniform_fit_param("logZ", -1, 3)
fit_info.add_uniform_fit_param("log_cloudtop_P", -0.99, 5)
fit_info.add_uniform_fit_param("error_multiple", 0, np.inf, 0.5, 5)
#Use Nested Sampling to do the fitting
result = retriever.run_emcee(bins, depths, errors, fit_info, plot_best=True)
plt.savefig("best_fit.png")
np.save("chain.npy", result.chain)
np.save("logl.npy", result.lnprobability)
fig = corner.corner(result.flatchain,
range=[0.99] * result.flatchain.shape[1],
labels=fit_info.fit_param_names)
fig.savefig("emcee_corner.png")
class PlatonWrapper():
"""Class object for running the platon atmospheric retrieval
software."""
def __init__(self):
"""Initialize the class object."""
self.ec2_id = ''
self.output_results = 'results.dat'
self.output_plot = 'corner.png'
self.retriever = Retriever()
self.ssh_file = ''
self.aws = False
self._configure_logging()
def _configure_logging(self):
"""Creates a log file that logs the execution of the script.
Log files are written to a ``logs/`` subdirectory within the
current working directory.
Returns
-------
start_time : obj
The start time of the script execution
"""
# Define save location
log_file = 'logs/{}.log'.format(
datetime.datetime.now().strftime('%Y-%m-%d-%H-%M'))
# Create the subdirectory if necessary
if not os.path.exists('logs/'):
os.mkdir('logs/')
# Make sure no other root handlers exist before configuring the logger
for handler in logging.root.handlers[:]:
logging.root.removeHandler(handler)
# Create the log file
logging.basicConfig(filename=log_file,
format='%(asctime)s %(levelname)s: %(message)s',
datefmt='%m/%d/%Y %H:%M:%S %p',
level=logging.INFO)
print('Log file initialized to {}'.format(log_file))
# Log environment information
logging.info('User: '******'System: ' + socket.gethostname())
logging.info('Python Version: ' + sys.version.replace('\n', ''))
logging.info('Python Executable Path: ' + sys.executable)
self.start_time = time.time()
def make_plot(self):
"""Create a corner plot that shows the results of the retrieval."""
print('Creating corner plot')
logging.info('Creating corner plot')
matplotlib.rcParams['text.usetex'] = False
if self.method == 'emcee':
fig = corner.corner(self.result.flatchain,
range=[0.99] * self.result.flatchain.shape[1],
labels=self.fit_info.fit_param_names)
elif self.method == 'multinest':
fig = corner.corner(self.result.samples,
weights=self.result.weights,
range=[0.99] * self.result.samples.shape[1],
labels=self.fit_info.fit_param_names)
# Save the results
self.output_plot = '{}_corner.png'.format(self.method)
fig.savefig(self.output_plot)
print('Corner plot saved to {}'.format(self.output_plot))
logging.info('Corner plot saved to {}'.format(self.output_plot))
def retrieve(self, method):
"""Perform the atmopsheric retrieval via the given method
Parameters
----------
method : str
The method by which to perform atmospheric retrievals. Can
either be ``emcee`` or ``multinest``."""
print('Performing atmopsheric retrievals via {}'.format(method))
logging.info('Performing atmopsheric retrievals via {}'.format(method))
# Ensure that the method parameter is valid
assert method in ['multinest',
'emcee'], 'Unrecognized method: {}'.format(method)
self.method = method
# For processing on AWS
if self.aws:
# Start or create an EC2 instance
instance, key, client = start_ec2(self.ssh_file, self.ec2_id)
# Build the environment on EC2 instance if necessary
if self.build_required:
build_environment(instance, key, client)
# Transfer object file to EC2
transfer_to_ec2(instance, key, client, 'pw.obj')
# Connect to the EC2 instance and run commands
command = './exoctk/exoctk/atmospheric_retrievals/exoctk-env-init.sh python exoctk/exoctk/atmospheric_retrievals/platon_wrapper.py {}'.format(
self.method)
client.connect(hostname=instance.public_dns_name,
username='******',
pkey=key)
stdin, stdout, stderr = client.exec_command(command)
output = stdout.read()
errors = stderr.read()
log_output(output)
log_output(errors)
# Trasfer output products from EC2 to user
if self.method == 'emcee':
transfer_from_ec2(instance, key, client, 'emcee_results.obj')
transfer_from_ec2(instance, key, client, 'emcee_corner.png')
elif self.method == 'multinest':
transfer_from_ec2(instance, key, client,
'multinest_results.dat')
transfer_from_ec2(instance, key, client,
'multinest_corner.png')
# Terminate or stop the EC2 instance
stop_ec2(self.ec2_id, instance)
# For processing locally
else:
if self.method == 'emcee':
self.result = self.retriever.run_emcee(self.bins, self.depths,
self.errors,
self.fit_info)
elif self.method == 'multinest':
self.result = self.retriever.run_multinest(self.bins,
self.depths,
self.errors,
self.fit_info,
plot_best=False)
_log_execution_time(self.start_time)
def save_results(self):
"""Save the results of the retrieval to an output file."""
print('Saving results')
logging.info('Saving results')
# Save the results
if self.method == 'multinest':
self.output_results = 'multinest_results.dat'
with open(self.output_results, 'w') as f:
f.write(str(self.result))
elif self.method == 'emcee':
self.output_results = 'emcee_results.obj'
with open(self.output_results, 'wb') as f:
pickle.dump(self.result, f)
print('Results file saved to {}'.format(self.output_results))
logging.info('Results file saved to {}'.format(self.output_results))
def set_parameters(self, params):
"""Set necessary parameters to perform the retrieval.
Required parameters include ``Rs``, ``Mp``, ``Rp``, and ``T``.
Optional parameters include ``logZ``, ``CO_ratio``,
``log_cloudtop_P``, ``log_scatt_factor``, ``scatt_slope``,
``error_multiple``, and ``T_star``.
Parameters
----------
params : str or dict
Either a path to a params file to use, or a dictionary of
parameters and their values for running the software.
See "Use" documentation for further details.
"""
print('Setting parameters: {}'.format(params))
logging.info('Setting parameters: {}'.format(params))
_validate_parameters(params)
_apply_factors(params)
self.params = params
self.fit_info = self.retriever.get_default_fit_info(**self.params)
def use_aws(self, ssh_file, ec2_id):
"""Sets appropriate parameters in order to perform processing
using an AWS EC2 instance.
Parameters
----------
ssh_file : str
The path to a public SSH key used to connect to the EC2
instance.
ec2_id : str
A template id that points to a pre-built EC2 instance.
"""
print('Using AWS for processing')
logging.info('Using AWS for processing')
self.ssh_file = ssh_file
self.ec2_id = ec2_id
# If the ec2_id is a template ID, then building the instance is required
if ec2_id.split('-')[0] == 'lt':
self.build_required = True
else:
self.build_required = False
# Write out object to file
with open('pw.obj', 'wb') as f:
pickle.dump(self, f)
print('Saved PlatonWrapper object to pw.obj')
logging.info('Saved PlatonWrapper object to pw.obj')
self.aws = True
class PlatonWrapper():
"""Class object for running the platon atmospheric retrieval
software."""
def __init__(self):
"""Initialize the class object."""
self.retriever = Retriever()
self.output_results = 'results.dat'
self.output_plot = 'corner.png'
def make_plot(self):
"""Create a corner plot that shows the results of the retrieval."""
if self.method == 'emcee':
fig = corner.corner(self.result.flatchain, range=[0.99] * self.result.flatchain.shape[1],
labels=self.fit_info.fit_param_names)
elif self.method == 'multinest':
fig = corner.corner(self.result.samples, weights=self.result.weights,
range=[0.99] * self.result.samples.shape[1],
labels=self.fit_info.fit_param_names)
# Save the results
self.output_plot = '{}_corner.png'.format(self.method)
fig.savefig(self.output_plot)
print('Corner plot saved to {}'.format(self.output_plot))
def retrieve_emcee(self):
"""Perform the atmopsheric retrieval via emcee."""
self.method = 'emcee'
self.result = self.retriever.run_emcee(self.bins, self.depths, self.errors, self.fit_info)
def retrieve_multinest(self):
"""Perform the atmopsheric retrieval via multinested sampling."""
self.method = 'multinest'
self.result = self.retriever.run_multinest(self.bins, self.depths, self.errors, self.fit_info, plot_best=False)
def save_results(self):
"""Save the results of the retrieval to an output file."""
# Save the results
self.output_results = '{}_results.dat'.format(self.method)
with open(self.output_results, 'w') as f:
f.write(str(self.result))
print('Results file saved to {}'.format(self.output_results))
def set_parameters(self, params):
"""Set necessary parameters to perform the retrieval.
Required parameters include ``Rs``, ``Mp``, ``Rp``, and ``T``.
Optional parameters include ``logZ``, ``CO_ratio``,
``log_cloudtop_P``, ``log_scatt_factor``, ``scatt_slope``,
``error_multiple``, and ``T_star``.
Parameters
----------
params : dict
A dictionary of parameters and their values for running the
software. See "Use" documentation for further details.
"""
_validate_parameters(params)
_apply_factors(params)
self.params = params
self.fit_info = self.retriever.get_default_fit_info(**self.params)
