def fit_multinest(self, n_live_points=1000, basename="chains/1-", verbose=True, overwrite=True, **kwargs):
self._mnest_basename = basename
# creates the directory for the output
# folder = os.path.abspath(os.path.dirname(self._mnest_basename))
# if not os.path.exists(self._mnest_basename):
# os.makedirs(self._mnest_basename)
if hasattr(self, "which"):
self.n_params = 9 + 6 * self.lc.n_planets
else:
self.n_params = 5 + 6 * self.lc.n_planets
pymultinest.run(
self.mnest_loglike,
self.mnest_prior,
self.n_params,
n_live_points=n_live_points,
outputfiles_basename=self._mnest_basename,
verbose=verbose,
**kwargs
)
self._make_samples()
def multinest(parameter_names, transform, loglikelihood, output_basename, **problem):
parameters = parameter_names
n_params = len(parameters)
def myprior(cube, ndim, nparams):
params = transform([cube[i] for i in range(ndim)])
for i in range(ndim):
cube[i] = params[i]
def myloglike(cube, ndim, nparams):
l = loglikelihood([cube[i] for i in range(ndim)])
return l
# run MultiNest
mn_args = dict(
outputfiles_basename = output_basename,
resume = problem.get('resume', False),
verbose = True,
n_live_points = problem.get('n_live_points', 400))
if 'seed' in problem:
mn_args['seed'] = problem['seed']
pymultinest.run(myloglike, myprior, n_params, **mn_args)
import json
# store name of parameters, always useful
with file('%sparams.json' % output_basename, 'w') as f:
json.dump(parameters, f, indent=2)
# analyse
a = pymultinest.Analyzer(n_params = n_params,
outputfiles_basename = output_basename)
s = a.get_stats()
with open('%sstats.json' % a.outputfiles_basename, mode='w') as f:
json.dump(s, f, indent=2)
return a
def lfit(xx, yy, error, bounds=[-10., 10., -10., 10.]):
# linear fit with nested sampling
# prevents seg fault in MultiNest
error[error == 0] = np.mean(error[error != 0])
def model_lin(params):
return params[0] + xx * params[1]
def myprior_lin(cube, ndim, n_params):
'''This transforms a unit cube into the dimensions of your prior space.'''
cube[0] = (bounds[1] - bounds[0]) * cube[0] + bounds[0]
cube[1] = (bounds[3] - bounds[2]) * cube[1] + bounds[2]
def myloglike_lin(cube, ndim, n_params):
loglike = -np.sum(((yy - model_lin(cube)) / error)**2)
return loglike / 2.
pymultinest.run(
myloglike_lin,
myprior_lin,
2,
resume=False,
sampling_efficiency=0.5,
evidence_tolerance=0.1,
)
# retrieves the data that has been written to hard drive
a = pymultinest.Analyzer(n_params=2)
posteriors = a.get_data()
stats = a.get_stats()
stats['marginals'] = get_stats(posteriors)
return stats, posteriors
def test():
test.prior_was_called = False
test.loglike_was_called = False
test.dumper_was_called = False
def myprior(cube, ndim, nparams):
for i in range(ndim):
cube[i] = cube[i] * 10 * math.pi
test.prior_was_called = True
def myloglike(cube, ndim, nparams):
chi = 1.
for i in range(ndim):
chi *= math.cos(cube[i] / 2.)
test.loglike_was_called = True
return math.pow(2. + chi, 5)
def mydumper(nSamples,nlive,nPar,
physLive,posterior,paramConstr,
maxLogLike,logZ,logZerr,nullcontext):
print("calling dumper")
test.dumper_was_called = True
# number of dimensions our problem has
parameters = ["x", "y"]
n_params = len(parameters)
# run MultiNest
pymultinest.run(myloglike, myprior, n_params,
resume = True, verbose = True,
dump_callback=mydumper)
assert test.prior_was_called
assert test.loglike_was_called
assert test.dumper_was_called
def perform_scan(self, run_tag=None, nlive=100, pymultinest_options=None):
""" When called creates the directories for and then performs the scan
run_tag: label of file in which the chains output is stored
nlive: number of live points to user during the scan. The default
value of 100 is chosen to speed up testing, but for actual runs
a larger value is recommended
pymultinest_options: Custom user inputs passed to multinest; must
be inserted in the form of a dictionary, as for the default below
"""
self.run_tag = run_tag
self.make_dirs_for_run(run_tag)
if not pymultinest_options:
# Set defaults
pymultinest_options = {
'importance_nested_sampling': False,
'resume': False,
'verbose': True,
'sampling_efficiency': 'model',
'init_MPI': False,
'evidence_tolerance': 0.5,
'const_efficiency_mode': False
}
else:
pass # Use passed parameters
# Run MultiNest
pymultinest.run(self.ll,
self.lp,
self.n_params,
outputfiles_basename=self.chains_dir_for_run,
n_live_points=nlive,
**pymultinest_options)
def RunMultinest():
def loglike(cube, ndim, nparams):
n_signal = events_gen_stat(cube)
ll = obs * log(n_signal) - n_signal - gammaln(obs + 1)
return sum(ll)
save_str = "cenns10_stat"
out_str = "multinest/" + save_str + "/" + save_str
json_str = "multinest/" + save_str + "/params.json"
# Run the sampler with CEvNS, BRN, and SS.
pymultinest.run(loglike,
prior_stat,
4,
outputfiles_basename=out_str,
resume=False,
verbose=True,
n_live_points=1000,
evidence_tolerance=0.5,
sampling_efficiency=0.8)
# Save the parameter names to a JSON file.
params_stat = [
"cevns_norm", "ss_norm", "BRN_prompt_norm", "BRN_delayed_norm"
]
json.dump(params_stat, open(json_str, 'w'))
def fit(self, verbose=False, n_live=400):
""" Fit the specified model to the input galaxy data.
Parameters
----------
verbose : bool - optional
Set to True to get progress updates from the sampler.
n_live : int - optional
Number of live points: reducing speeds up the code but may
lead to unreliable results.
"""
if "lnz" in list(self.results):
print("Fitting not performed as results have already been" +
" loaded from " + self.fname[:-1] + ".h5. To start" +
" over delete this file or change run.\n")
return
print("\nBagpipes: fitting object " + self.galaxy.ID + "\n")
start_time = time.time()
with warnings.catch_warnings():
warnings.simplefilter("ignore")
pmn.run(self.fitted_model.lnlike,
self.fitted_model.prior.transform,
self.fitted_model.ndim,
importance_nested_sampling=False,
verbose=verbose,
sampling_efficiency="model",
n_live_points=n_live,
outputfiles_basename=self.fname)
runtime = time.time() - start_time
print("\nCompleted in " + str("%.1f" % runtime) + " seconds.\n")
# Load MultiNest outputs and save basic quantities to file.
samples2d = np.loadtxt(self.fname + "post_equal_weights.dat")[:, :-1]
lnz_line = open(self.fname + "stats.dat").readline().split()
self.results["samples2d"] = samples2d
self.results["lnz"] = float(lnz_line[-3])
self.results["lnz_err"] = float(lnz_line[-1])
self.results["median"] = np.median(samples2d, axis=0)
self.results["conf_int"] = np.percentile(samples2d, (16, 84), axis=0)
# Save re-formatted outputs as HDF5 and remove MultiNest output.
with warnings.catch_warnings():
warnings.simplefilter("ignore")
dd.io.save(self.fname[:-1] + ".h5", self.results)
os.system("rm " + self.fname + "*")
# Create a posterior object to hold the results of the fit.
self.posterior = posterior(self.galaxy, run=self.run)
self._print_results()
def main():
# Begin timing the estimation process
start_time = time.time()
# Run the MultiNest software
pmn.run(binary_logit_log_likelihood, uniform_prior, num_dimensions,
outputfiles_basename=relative_output_folder,
n_live_points=num_live_points,
sampling_efficiency=desired_sampling_efficiency,
log_zero=-1e200,
mode_tolerance=-1e180,
null_log_evidence=-1e180,
resume=False, verbose=True, init_MPI=False)
# Record the ending time of the estimation process
end_time = time.time()
tot_minutes = (end_time - start_time) / 60.0
# Save the parameter names
with open(relative_output_folder + "parameter_names.json", 'wb') as f:
json.dump(explanatory_vars, f)
# Save the number of live points used as the total estimation time
model_run_params = {"n_live_points": num_live_points,
"sampling_efficiency": desired_sampling_efficiency,
"estimation_minutes": tot_minutes}
with open(relative_output_folder + "model_run_parameters.json", "wb") as f:
json.dump(model_run_params, f)
# Print a report on how long the estimation process took
print "Estimation process took {:.2f} minutes".format(tot_minutes)
def run(self):
pymultinest.run(self.mod_file.loglikelihood,
self.mod_file.prior,
self.mod_file.n_pars,
n_clustering_params=None,
wrapped_params=None,
importance_nested_sampling=True,
multimodal=True,
const_efficiency_mode=False,
n_live_points=self.n_live,
evidence_tolerance=self.tolerance,
sampling_efficiency=0.8,
n_iter_before_update=100,
null_log_evidence=-1e+90,
max_modes=100,
mode_tolerance=-1e+90,
outputfiles_basename=self.model_name + "/chains/",
seed=-1,
verbose=True,
resume=self.resume,
context=0,
write_output=True,
log_zero=-1e+100,
max_iter=self.max_iterations,
init_MPI=False,
dump_callback=None)
def RunMultinestNull():
def loglike(cube, ndim, nparams):
n_signal = events_gen_stat_null(cube)
ll = cut_crit*nan_to_num(obs * log(n_signal) - n_signal - gammaln(obs+1)) \
- (cube[0] - 1)**2 / (2 * ss_error**2) \
- (cube[1] - 1)**2 / (2 * 0.3**2) \
- (cube[2] - 1)**2 / (2)
return sum(nan_to_num(ll))
save_str = "cenns10_stat_null_llConstraint"
out_str = "multinest/" + save_str + "/" + save_str
json_str = "multinest/" + save_str + "/params.json"
# Run the sampler with just BRN, and SS.
pymultinest.run(loglike,
prior_stat_null,
3,
outputfiles_basename=out_str,
resume=False,
verbose=True,
n_live_points=1000,
evidence_tolerance=0.1,
sampling_efficiency=0.8)
# Save the parameter names to a JSON file.
params_stat_null = ["ss_norm", "BRN_prompt_norm", "BRN_delayed_norm"]
json.dump(params_stat_null, open(json_str, 'w'))
def fit_multinest(self,
n_live_points=1000,
basename='chains/1-',
verbose=True,
overwrite=True,
**kwargs):
self._mnest_basename = basename
#creates the directory for the output
#folder = os.path.abspath(os.path.dirname(self._mnest_basename))
#if not os.path.exists(self._mnest_basename):
# os.makedirs(self._mnest_basename)
if hasattr(self, 'which'): self.n_params = 9 + 6 * self.lc.n_planets
else: self.n_params = 5 + 6 * self.lc.n_planets
pymultinest.run(self.mnest_loglike,
self.mnest_prior,
self.n_params,
n_live_points=n_live_points,
outputfiles_basename=self._mnest_basename,
verbose=verbose,
**kwargs)
self._make_samples()
def RunMultinestNull():
def marginalizedLikelihood(cube, ndim, nparams):
def likelihood(s1, s2, s3):
signal = events_gen_null([cube[0], cube[1], cube[2], s1, s2, s3])
return np.prod(pois(
signal, obs)) * systGaus(s1) * systGaus(s2) * systGaus(s3)
ll = log(nquad(likelihood, [[0.0, 1.0], [0.0, 1.0], [0.5, 1.0]])[0])
print(ll)
return ll
save_str = "syst_marginalized_null"
out_str = "multinest/" + save_str + "/" + save_str
json_str = "multinest/" + save_str + "/params.json"
# Run the sampler with just BRN, and SS.
pymultinest.run(marginalizedLikelihood,
prior_null,
6,
outputfiles_basename=out_str,
resume=False,
verbose=True,
n_live_points=2000,
evidence_tolerance=0.5,
sampling_efficiency=0.8)
# Save the parameter names to a JSON file.
params_null = ["ss_norm", "BRN_prompt_norm", "BRN_delayed_norm"]
json.dump(params_null, open(json_str, 'w'))
def run(self, outputfiles_basename, resume=False, verbose=True):
'''
:param outputfiles_basename:
:param resume:
:param verbose:
:return:
'''
if c_lib_link is not None:
pymultinest.run(self.log_like_c,
self.prior,
self.n_params,
outputfiles_basename=outputfiles_basename,
resume=resume,
verbose=verbose)
else:
pymultinest.run(self.log_like,
self.prior,
self.n_params,
outputfiles_basename=outputfiles_basename,
resume=resume,
verbose=verbose)
#a1 = pymultinest.Analyzer(outputfiles_basename=outputfiles_basename, n_params = self.n_params)
a1 = Analyzer(outputfiles_basename=outputfiles_basename,
n_params=self.n_params)
return a1
def multinest(optimizer, nprocs=1):
# number of dimensions our problem has
parameters = ["{0}".format(i)
for i in range(len(optimizer.params.get_all(True)))]
nparams = len(parameters)
if not os.path.exists('chains'):
os.mkdir('chains')
def lnprior(cube, ndim, nparams):
theta = np.array([cube[i] for i in range(ndim)])
for i in range(len(optimizer.params.get_all(True))):
param = optimizer.params.get_all(True)[i]
if "mass_" in param.name:
theta[i] = 10 ** (theta[i] * 8 - 9)
elif "radius_" in param.name:
theta[i] = 10 ** (theta[i] * 4 - 4)
elif "flux_" in param.name:
theta[i] = 10 ** (theta[i] * 4 - 4)
elif "a_" in param.name:
theta[i] = 10 ** (theta[i] * 2 - 2)
elif "e_" in param.name:
theta[i] = 10 ** (theta[i] * 3 - 3)
elif "inc_" in param.name:
theta[i] *= 2.0 * np.pi
elif "om_" in param.name:
theta[i] = 2.0 * np.pi * 10 ** (theta[i] * 2 - 2)
elif "ln_" in param.name:
theta[i] = 2.0 * np.pi * 10 ** (theta[i] * 8 - 8)
elif "ma_" in param.name:
theta[i] = 2.0 * np.pi * 10 ** (theta[i] * 2 - 2)
for i in range(ndim):
cube[i] = theta[i]
def lnlike(cube, ndim, nparams):
theta = np.array([cube[i] for i in range(ndim)])
optimizer.params.update(theta)
mod_flux, mod_rv = optimizer.model(nprocs)
flnl = -(0.5 * ((mod_flux - optimizer.photo_data[1]) /
optimizer.photo_data[2]) ** 2)
rvlnl = -(0.5 * ((mod_rv - optimizer.rv_data[1]) /
optimizer.rv_data[2]) ** 2)
tlnl = np.sum(flnl) + np.sum(rvlnl)
nobj = np.append(np.sum(flnl) + np.sum(rvlnl), theta)
optimizer.chain = np.vstack([optimizer.chain, nobj])
if tlnl > optimizer.maxlnp:
optimizer.iterout(tlnl, theta, mod_flux)
return np.sum(flnl) + np.sum(rvlnl)
# run MultiNest
pymultinest.run(lnlike, lnprior, nparams, n_live_points=1000)
def fit_without_background_uncertainty(events_generator,
n_params,
n_bg,
n_obs,
sigma,
prior,
out_put_dir,
resume=False,
verbose=True,
n_live_points=1500,
evidence_tolerance=0.1,
sampling_efficiency=0.3,
**kwargs):
"""
fitting data using provided loglikelihood,
assumming no uncertainty on background,
the default of each parameter ranges from 0 to 1 uniformly,
use prior to modify the range or distribution of parameters
:param events_generator: functions to generate predicted number of events
:param n_params: number of parameters to be fitted
:param n_bg: background data, 1d array
:param n_obs: experiment observed data, 1d array
:param sigma: systemmatic uncertainty on signal
:param prior: prior for each parameters
:param out_put_dir: out put directories
:param resume: multinest parameter, default is False
:param verbose: multinest parameter, default is True
:param n_live_points: multinest parameter, default is 1500
:param evidence_tolerance: multinest parameter, default is 0.1
:param sampling_efficiency: multinest parameter, default is 0.3
:param kwargs other parameters for multinest
"""
# pymultinest requires ndim and nparams in loglikelihood and prior, it seems we don't need it here
def lgl(cube, ndim, nparams):
n_signal = events_generator(cube)
likelihood = np.zeros(n_obs.shape[0])
for i in range(n_obs.shape[0]):
likelihood[i] = quad(
lambda a: _poisson(n_obs[i], (1 + a) * n_signal[i] + n_bg[i]) *
_gaussian(a, 0, sigma), -3 * sigma, 3 * sigma)[0]
prod_like = np.prod(likelihood)
return np.log(prod_like) if prod_like > 0 else -np.inf
def prr(cube, ndim, nparams):
prior(cube)
pymultinest.run(lgl,
prr,
n_params,
outputfiles_basename=out_put_dir + '_',
resume=resume,
verbose=verbose,
n_live_points=n_live_points,
evidence_tolerance=evidence_tolerance,
sampling_efficiency=sampling_efficiency,
**kwargs)
def multinestevidence(mean,cov,model):
"""
Input prior over paramters and black box function for evaluating ln liklihood
based on chi2 from image/model and calcualte evidence for given model
in preferebly the exact same way to emcee to allow comparison
"""
prior1 = prior(mean,cov)
pymultinest.run(lnlike, prior1.pdf, n_params,
resume = True, verbose = True)
def run_scan(self):
if os.path.exists(self.chain_path):
shutil.rmtree(self.chain_path)
os.makedirs(self.chain_path)
os.chdir(self.chain_path)
pymultinest.run(self.likelihood_val, self.prior, 3, **self.pnest_pars)
os.chdir(path)
return
def multinestrun(LogLikelihood,my_prior,n_dims,ES,n_params,**kwargs):
'''
Author: yuanfang
Data: 2017.11.17
Run multinest
'''
yy=raw_input('running in yuanfang multinest function...')
#initialize the cube
cube=[0.0]*n_params
# def my_prior(cube,ndim,n_params):
# '''
# generate prior for multinest algorithm
# return cube whose 1~n_dim is the input parameter
# 2017.11.19
# '''
# for i,name in enumerate(ES.InPar):
# if ES.InputPar[name][1].lower() == 'flat':
# min = float(ES.InputPar[name][2])
# max = float(ES.InputPar[name][3])
# cube[i] = random() * (max - min) + min
# elif ES.InputPar[name][1].lower() == 'log':
# import math
# min = math.log10(float(ES.InputPar[name][2]))
# max = math.log10(float(ES.InputPar[name][3]))
# cube[i] = 10.0**(random()*(max - min) + min )
# else:
# sf.ErrorStop( 'Not ready. Only "flat" and "log" prior can be used.' )
import pymultinest
pymultinest.run(
LogLikelihood = LogLikelihood,
Prior = my_prior,
n_dims = len(ES.InPar),
n_params = n_params,
seed = ES.getRandomSeed(),
outputfiles_basename = ES.MNOutputFile,
n_live_points = kwargs.pop('n_live_points',2*n_dims),
n_clustering_params = kwargs.pop('n_clustering_params',2),
multimodal = kwargs.pop('multimodal',True),
const_efficiency_mode= kwargs.pop('const_efficiency_mode', False), # set to false by default
evidence_tolerance = kwargs.pop('evidence_tolerance',1),
sampling_efficiency = kwargs.pop('sampling_efficiency',1),
n_iter_before_update = kwargs.pop('n_iter_before_update',ES.getPointNum()),
null_log_evidence = kwargs.pop('null_log_evidence',-1e+100),
max_modes = kwargs.pop('max_modes',5),
verbose = kwargs.pop('verbose',True),
resume = kwargs.pop('resume',False), # !!!!!!!!
context = kwargs.pop('context',0),
importance_nested_sampling = kwargs.pop('importance_nested_sampling',True)
)
json.dump(ES.InPar, open(ES.MNOutputFile+'input.json', 'w')) # save parameter names
def run_multinest(rperp, sigmag, invcov, splashback, outfile):
def Prior(cube, ndim, nparams):
# Sigma Values are from Chang 2018 Table 2. Each sigma is half a prior range
cube[0] = uniform(-0.92, -0.22, cube[0]) # log(alpha)
cube[1] = uniform(0.28, 1.28, cube[1]) # log(beta)
cube[2] = uniform(-0.4, 1.5, cube[2]) # log(gamma)
cube[3] = uniform(-1.02, -0.62, cube[3]) # r_s
cube[4] = uniform(-1., 19.,
cube[4]) # r_t #17 is probably better than 19
cube[5] = uniform(-1.4, -0.9, cube[5]) # rho_0
cube[6] = uniform(0.8, 2.4, cube[6]) # rho_s # 1.6 \pm 0.8
cube[7] = uniform(1.17, 1.77, cube[7]) # s_e
cube[8] = uniform(-0.9, 0., cube[8]) # ln(c_mis)
cube[9] = uniform(0.11, 0.7, cube[9]) # f_mis
def Loglike(cube, ndim, nparams):
# Read in parameters
log_alpha = cube[0]
log_beta = cube[1]
log_gamma = cube[2]
r_s = cube[3]
r_t = cube[4]
rho_0 = cube[5]
rho_s = cube[6]
se = cube[7]
ln_mis = cube[8]
f_mis = cube[9]
params = [
log_alpha, log_beta, log_gamma, r_s, r_t, rho_0, rho_s, se,
ln_mis, f_mis
]
# Calculate likelihood
sig_m = Sigmag(rperp, z, params, h0, splashback)
vec = sig_m - sigmag
likelihood = -0.5 * np.matmul(np.matmul(vec, invcov), vec.T)
# Calculate prior
prior = -0.5 * (-1.13 - ln_mis)**2 / 0.22**2 - 0.5 * (
log_alpha - np.log10(0.19))**2 / 0.4**2 - 0.5 * (
log_beta - np.log10(6.0))**2 / 0.4**2 - 0.5 * (
log_gamma - np.log10(4.0))**2 / 0.4**2 - 0.5 * (
f_mis - 0.22)**2 / 0.11**2
#prior = 0.
# Total probability
tot = likelihood + prior
return tot
# Run Multinest
mult.run(Loglike,
Prior,
10,
outputfiles_basename=outfile,
verbose=False)
def main():
cube = [ 0.9, 0.5, 0.1 ] # initial values not used
ndim = len(cube)
nparams = len(cube)
os.chdir('/home/jordi/allst/sample')
pm.run(F_calc_Likelihood4stmd, F_allpriors, nparams,importance_nested_sampling = False,
resume = False, verbose = True, n_live_points=32, outputfiles_basename="DMco_",
sampling_efficiency=0.02,const_efficiency_mode=True,init_MPI=False)
def RunMultinest():
def marginalizedLikelihood(cube, ndim, nparams):
def likelihood(s1, s2, s3, s4, s5):
signal = events_gen(
[cube[0], cube[1], cube[2], cube[3], s1, s2, s3, s4, s5])
return np.prod(pois(signal, obs)) * systGaus(s1) * systGaus(
s2) * systGaus(s3) * systGaus(s4) * systGaus(s5)
ll = log(
nquad(likelihood, [[0.0, 1.0], [0.5, 1.0], [0.0, 1.0], [0.0, 1.0],
[0.5, 1.0]])[0])
print(ll)
return ll
npoints = 100
rvs1 = np.random.uniform(0.0, 1.0, npoints)
rvs2 = np.random.uniform(0.5, 1.0, npoints)
rvs3 = np.random.uniform(0.0, 1.0, npoints)
rvs4 = np.random.uniform(0.0, 1.0, npoints)
rvs5 = np.random.uniform(0.5, 1.0, npoints)
volume = 0.5**2 / npoints
def mcLikelihood(cube, ndim, nparams):
def likelihood(s1, s2, s3, s4, s5):
signal = events_gen(
[cube[0], cube[1], cube[2], cube[3], s1, s2, s3, s4, s5])
return np.sum(log(pois(signal, obs) + 1.0)) + log(
systGaus(s1) * systGaus(s2) * systGaus(s3) * systGaus(s4) *
systGaus(s5))
ll = np.sum([
likelihood(rvs1[i], rvs2[i], rvs3[i], rvs4[i], rvs5[i])
for i in range(0, 50)
]) * volume
return ll
save_str = "syst_marginalized"
out_str = "multinest/" + save_str + "/" + save_str
json_str = "multinest/" + save_str + "/params.json"
# Run the sampler with CEvNS, BRN, and SS.
pymultinest.run(mcLikelihood,
prior,
9,
outputfiles_basename=out_str,
resume=False,
verbose=True,
n_live_points=2000,
evidence_tolerance=0.5,
sampling_efficiency=0.8)
# Save the parameter names to a JSON file.
params = ["cevns_norm", "ss_norm", "BRN_prompt_norm", "BRN_delayed_norm"]
json.dump(params, open(json_str, 'w'))
def run(self, outputfiles_basename, resume=False, verbose=True):
pymultinest.run(self.log_like,
self.prior,
self.n_params,
outputfiles_basename=outputfiles_basename,
resume=resume,
verbose=verbose)
a1 = Analyzer(outputfiles_basename=outputfiles_basename,
n_params=self.n_params)
data = [self.energyc, self.lag, self.lag_errl, self.lag_errh]
return Fit_plot(a1, self.parameters, data, self.model)
def my_curvfit(function,
prior,
parameters,
x,
y,
yerr1,
yerr2=None,
savedir=None,
done=True):
n_params = len(parameters)
if yerr2 is not None:
yerr2 = np.abs(yerr2)
x = np.array(x)
y = np.array(y)
yerr1 = np.abs(yerr1)
if savedir is None:
savedir = '/home/laojin/software/my_python/curvfit/'
if os.path.exists(savedir) == False:
os.makedirs(savedir)
else:
if os.path.exists(savedir) == False:
os.makedirs(savedir)
def loglike(cube, ndim, nparams):
ymodel = function(x, cube)
if yerr2 is not None:
err = []
for index, value in enumerate(ymodel - y):
if value > 0:
err.append(yerr2[index])
else:
err.append(yerr1[index])
err = np.array(err)
else:
err = yerr1
loglikelihood = (-0.5 * ((ymodel - y) / err)**2).sum()
return loglikelihood
if ((os.path.exists(savedir + 'spectrum_params.json') == False) or (done)):
pymultinest.run(loglike,
prior,
n_params,
outputfiles_basename=savedir + 'spectrum_',
resume=False,
verbose=True)
json.dump(parameters, open(savedir + 'spectrum_params.json', 'w'))
a = pymultinest.Analyzer(outputfiles_basename=savedir + 'spectrum0_',
n_params=n_params)
return a
def __call__(self, generator, prior, nparams):
def lgl(cube, ndim, nparams):
n_pred = generator(cube)
return self.loglike(self.n_obs, self.n_bg, n_pred)
def prr(cube, ndim, nparams):
prior(cube)
pymultinest.run(lgl,
prr,
nparams,
outputfiles_basename=self.output_dir + '_',
**self.kwargs)
def run_multinest(posterior, save_file):
"""Uses multinest sampler to calculate evidence instead of emceee
cmd is bash command to call lagrange_cpp
posterior is posterior class, should have methods prior and lik
save_file is path to save. will resume from file if arround"""
# checks
# if path exsissts
if not os.path.exists(save_file) and mpi.COMM_WORLD.rank == 0:
os.mkdir(save_file)
assert hasattr(posterior, 'prior') and hasattr(posterior, 'lik'), 'must have prior and lik methods'
# run sampler
pymultinest.run(posterior.lik, posterior.prior, posterior.get_dim(),
outputfiles_basename=save_file)
def sample(self):
self.pars = np.zeros(self.nParams)
# run MultiNest
pymultinest.run(self.logLike,
self.prior,
self.nParams,
importance_nested_sampling=False,
resume=False,
verbose=True,
sampling_efficiency='parameter',
n_live_points=100)
def generate(lmod_pars, lparams, lphoto_data, lrv_data, lncores, lfname):
global mod_pars, params, photo_data, rv_data, ncores, fname
mod_pars, params, photo_data, rv_data, ncores, fname = \
lmod_pars, lparams, lphoto_data, lrv_data, lncores, lfname
# number of dimensions our problem has
parameters = ["{0}".format(i) for i in range(mod_pars[0] * 5 + (mod_pars[0] - 1) * 6)]
nparams = len(parameters)
# make sure the output directories exist
if not os.path.exists("./output/{0}/multinest".format(fname)):
os.makedirs(os.path.join("./", "output", "{0}".format(fname), "multinest"))
if not os.path.exists("./output/{0}/plots".format(fname)):
os.makedirs(os.path.join("./", "output", "{0}".format(fname), "plots"))
if not os.path.exists("chains"): os.makedirs("chains")
# we want to see some output while it is running
progress_plot = pymultinest.ProgressPlotter(n_params=nparams,
outputfiles_basename='output/{0}/multinest/'.format(fname))
progress_plot.start()
# progress_print = pymultinest.ProgressPrinter(n_params=nparams, outputfiles_basename='output/{0}/multinest/'.format(fname))
# progress_print.start()
# run MultiNest
pymultinest.run(lnlike, lnprior, nparams, outputfiles_basename=u'./output/{0}/multinest/'.format(fname),
resume=True, verbose=True,
sampling_efficiency='parameter', n_live_points=1000)
# run has completed
progress_plot.stop()
# progress_print.stop()
json.dump(parameters, open('./output/{0}/multinest/params.json'.format(fname), 'w')) # save parameter names
# plot the distribution of a posteriori possible models
plt.figure()
plt.plot(photo_data[0], photo_data[1], '+ ', color='red', label='data')
a = pymultinest.Analyzer(outputfiles_basename="./output/{0}/reports/".format(fname), n_params=nparams)
for theta in a.get_equal_weighted_posterior()[::100, :-1]:
params = utilfuncs.split_parameters(theta, mod_pars[0])
mod_flux, mod_rv = utilfuncs.model(mod_pars, params, photo_data[0], rv_data[0])
plt.plot(photo_data[0], mod_flux, '-', color='blue', alpha=0.3, label='data')
utilfuncs.report_as_input(params, fname)
plt.savefig('./output/{0}/plots/posterior.pdf'.format(fname))
plt.close()
def perform_scan_multinest(self, chains_dir, nlive=100):
""" Perform a scan with MultiNest
"""
self.make_dirs([chains_dir])
n_params = len(self.floated_params)
pymultinest_options = {'importance_nested_sampling': False,
'resume': False, 'verbose': True,
'sampling_efficiency': 'model',
'init_MPI': False, 'evidence_tolerance': 0.5,
'const_efficiency_mode': False}
pymultinest.run(self.ll, self.prior_cube, n_params,
outputfiles_basename=chains_dir,
n_live_points=nlive, **pymultinest_options)
def run_Multinest(self,Nthreads,output_filename='temporary_model'): ''' This function runs Multinest using the self.my_prior priors, which need to be appropriately set to a Multinest-type of priors. Parameters Nthreads: int Number of threads per Multinest instance used by Galario to compute the model visibilities Output_filename: string Multinest output base file names. See pymultinest docs for how to read and analyze those. ''' n_params = 10 from galario import double double.threads(Nthreads) pymultinest.run(self.loglike_multinest, self.my_prior, n_params, outputfiles_basename=output_filename, resume = True, verbose = True,sampling_efficiency='model')
def _fit(self,
model: AbstractPriorModel,
analysis,
log_likelihood_cap=None) -> res.Result:
"""
Fit a model using MultiNest and the Analysis class which contains the data and returns the log likelihood from
instances of the model, which the `NonLinearSearch` seeks to maximize.
Parameters
----------
model : ModelMapper
The model which generates instances for different points in parameter space.
analysis : Analysis
Contains the data and the log likelihood function which fits an instance of the model to the data, returning
the log likelihood the `NonLinearSearch` maximizes.
Returns
-------
A result object comprising the Samples object that includes the maximum log likelihood instance and full
set of accepted ssamples of the fit.
"""
# noinspection PyUnusedLocal
def prior(cube, ndim, nparams):
# NEVER EVER REFACTOR THIS LINE! Haha.
phys_cube = model.vector_from_unit_vector(unit_vector=cube)
for i in range(len(phys_cube)):
cube[i] = phys_cube[i]
return cube
fitness_function = self.fitness_function_from_model_and_analysis(
model=model, analysis=analysis)
import pymultinest
logger.info("Beginning MultiNest non-linear search. ")
pymultinest.run(fitness_function,
prior,
model.prior_count,
outputfiles_basename="{}/multinest".format(
self.paths.path),
verbose=not self.silence,
**self.config_dict_search)
self.copy_from_sym()
def test():
def myprior(cube, ndim, nparams):
for i in range(ndim):
cube[i] = cube[i] * 10 * math.pi
def myloglike(cube, ndim, nparams):
chi = 1.
for i in range(ndim):
chi *= math.cos(cube[i] / 2.)
return math.pow(2. + chi, 5)
# number of dimensions our problem has
parameters = ["x", "y"]
n_params = len(parameters)
# run MultiNest
pymultinest.run(myloglike, myprior, n_params, resume=True, verbose=True)
def sample(self, nlive = 500, ceff = False, efr = 0.2, resume = False, doplot = False, sample=True): ''' Function to begin sampling with model defined in Candidate File. nlive - number of live points for multinest (default = 500) ceff - use constant efficient mode (default = False) efr - efficiency rate (default = 0.2) resume - resume sampling if restarted (default = False) doplot - make plots after sampling (default = False) ''' if(sample == True): pymultinest.run(self.GaussGPULikeWrap, self.MNprior, self.Cand.n_dims, n_params = self.Cand.n_params, importance_nested_sampling = False, resume = resume, verbose = True, sampling_efficiency = efr, multimodal=False, const_efficiency_mode = ceff, n_live_points = nlive, outputfiles_basename=self.ChainRoot, wrapped_params=self.Cand.wrapped) self.loadChains() if(doplot == True): self.plotResult()
def main():
"""
"""
# Set up MPI variables
world=MPI.COMM_WORLD
rank=world.rank
size=world.size
master = rank==0
if master:
print "Runtime parameters"
pprint.pprint(rp)
time.sleep(2)
if not os.path.exists(rp["outdir"]):
try:
os.mkdir(rp["outdir"])
except:
pass
n_params = rp["nc_fit"] + 3
#progress = pymultinest.ProgressPlotter(n_params=n_params, interval_ms=10000,
# outputfiles_basename=rp["outputfiles_basename"])
#progress.start()
pymultinest.run(loglike, logprior, n_params, resume=False, verbose=True,
multimodal=rp["multimodal"], max_modes=rp["max_modes"], write_output=True,
n_live_points=rp["n_live_points"],
evidence_tolerance=rp["evidence_tolerance"],
mode_tolerance=rp["mode_tolerance"],
seed=rp["seed"],
max_iter=rp["max_iter"],
importance_nested_sampling=rp["do_ins"],
outputfiles_basename=rp["outputfiles_basename"],\
init_MPI=False)
if master:
# Copy the config.ini file to the output dir
shutil.copy(param_file,rp["outdir"])
#progress.stop()
return 0
def main():
"""
"""
# Set up MPI variables
world=MPI.COMM_WORLD
rank=world.rank
size=world.size
master = rank==0
if master:
print("Runtime parameters")
pprint.pprint(rp)
time.sleep(2)
if not os.path.exists(rp["outdir"]):
try:
os.mkdir(rp["outdir"])
except:
pass
n_params = rp["nc_fit"] + 3
#progress = pymultinest.ProgressPlotter(n_params=n_params, interval_ms=10000,
# outputfiles_basename=rp["outputfiles_basename"])
#progress.start()
pymultinest.run(loglike, logprior, n_params, resume=False, verbose=True,
multimodal=rp["multimodal"], max_modes=rp["max_modes"], write_output=True,
n_live_points=rp["n_live_points"],
evidence_tolerance=rp["evidence_tolerance"],
mode_tolerance=rp["mode_tolerance"],
seed=rp["seed"],
max_iter=rp["max_iter"],
importance_nested_sampling=rp["do_ins"],
outputfiles_basename=rp["outputfiles_basename"],\
init_MPI=False)
if master:
# Copy the config.ini file to the output dir
shutil.copy(param_file,rp["outdir"])
#progress.stop()
return 0
def main():
"""
"""
if not os.path.exists(outdir): DIR(outdir)
n_params = 6
pymultinest.run(myloglike,myprior,n_params,resume=False,verbose=True,\
multimodal=multimodal,max_modes=max_modes,write_output=True,\
n_live_points=n_live_points,\
evidence_tolerance=evidence_tolerance,\
mode_tolerance=-1e90,seed=seed,max_iter=max_iter,\
importance_nested_sampling=do_INS,\
outputfiles_basename=os.path.join(outdir,outstem),init_MPI=False)
#contour_plot.contourTri(pylab.loadtxt('chains_test/1-post_equal_weights.dat'),line=True,outfile='chains_test/test.png',col=('red','blue'),labels=parameters,binsize=50,truth=plotTruth,ranges=plotRanges,autoscale=True)
return 0
def run(self, clean_up=None, **kwargs):
if clean_up is None:
if self.run_dir is None:
clean_up = True
else:
clean_up = False
if self.run_dir is None:
run_dir = tempfile.mkdtemp()
else:
run_dir = self.run_dir
basename = self.prepare_fit_directory(run_dir, self.prefix)
start_time = time.time()
logger.info('Starting fit in {0} with prefix {1}'.format(
run_dir, self.prefix))
pymultinest.run(self.likelihood.multinest_evaluate,
self.priors.prior_transform,
self.n_params,
outputfiles_basename='{0}_'.format(basename),
**kwargs)
logger.info("Fit finished - took {0:.2f} s".format(time.time() -
start_time))
fitted_parameter_names = [
item for item in self.likelihood.param_names
if not self.likelihood.fixed[item]
]
self.result = MultiNestResult.from_multinest_basename(
basename, fitted_parameter_names)
if clean_up:
logger.info("Cleaning up - deleting {0}".format(run_dir))
shutil.rmtree(run_dir)
else:
logger.info("Multinest files can be found in {0}".format(run_dir))
self.likelihood.parameters[~self.likelihood.fixed_mask()] = (
self.result.median.values)
return self.result
def find_maximum_3(x, y, y_sd, basename, plotit=False):
"""
guass fit using multinest
"""
A_lim = [0.8 * y.max(), 1.2 * y.max()]
X0_lim = [x[0], x[-1]]
S_lim = [0.5 * (x[-1]**2 - x[0]**2), 3 * (x[-1]**2 - x[0]**2)]
def Prior(cube, ndim, nparams):
cube[0] = cube[0] * (A_lim[1] - A_lim[0]) + A_lim[0]
cube[1] = cube[1] * (X0_lim[1] - X0_lim[0]) + X0_lim[0]
cube[2] = cube[2] * (S_lim[1] - S_lim[0]) + S_lim[0]
def LogLikelihood(cube, ndim, nparams):
fit = _gauss(cube, x)
chisq = np.sum((fit - y)**2 / y_sd**2)
return -chisq / 2.0
nparams = 3
pymultinest.run(LogLikelihood,
Prior,
nparams,
resume=False,
outputfiles_basename=basename)
result = pymultinest.analyse.Analyzer(nparams,
outputfiles_basename=basename)
samples = result.get_equal_weighted_posterior()
pks = samples[:, 1]
pk = np.mean(pks)
pk_err = np.std(pks)
if plotit:
f, ax = plt.subplots(1, 2)
ax[0].errorbar(x, y, yerr=y_sd, fmt='.', color='C0')
ax[0].axvspan(pk - pk_err, pk + pk_err, color='C1', alpha=0.3)
my_hist(ax[1], pks)
plt.show()
return pk, pk_err
def pelim(infile,doopt,spar,sopt,gpar,gopt,gpix, psfpar,\
crval=[0.,0.],cdelt=[1.,1.],crpix=[0.,0.],noise=1.0, comps=None):
if len(np.ravel(spar))==12:
snames = np.array(['SX1','SY1','SF1','SW1','SE1','SPA1']) #,'SX2','SY2','SF2','SW2','SE2','SPA2'
else:
snames = np.array(['SX1','SY1','SF1','SW1','SE1','SPA1'])
gnames = ['GX','GY','GB','GE','GPA','GSM','GSA']
global b
b=getdata(infile)
b=b[0][0] if b.ndim==4 else b
global lin, olin, x0
lin = np.append(np.ravel(spar),np.ravel(gpar))
olin = np.append(np.ravel(sopt),np.ravel(gopt))
x0 = lin[olin]
global parameters
both_names = np.append(np.ravel(snames),np.ravel(gnames))
parameters = both_names[olin]
global last_retval
last_retval = None
if doopt:
args=(spar,sopt,gpar,gopt,gpix, psfpar,crval,cdelt,crpix,\
b,False,noise,False, comps, lin)
n_params = len(parameters)
print ('n_params:', n_params)
datafile = 'out/1115'
print (parameters)
# run MultiNest
pymultinest.run(loglike, prior, n_params, outputfiles_basename=datafile + '_1_', resume = False, verbose = True)
json.dump(parameters.tolist(), open(datafile + '_1_params.json', 'w')) # save parameter names
#xopt= lbfgs(optim_func, x0, args=args, approx_grad = 1)
#xopt= fmin(optim_func, x0, args=args, maxfun=10000)
else:
optim_func(x0,spar,sopt,gpar,gopt,gpix, psfpar,crval,cdelt,crpix,\
b,False,noise,True, comps, lin)
ospar,ogpar = replace_x0 (spar,gpar,sopt,gopt,xopt if doopt else x0)
def run_nested(spec, model, basename='run/test_run'):
pymultinest.run(
model.loglikelihood,
model.prior_transform,
model.n_params,
outputfiles_basename=basename,
resume=False,
verbose=True,
evidence_tolerance=0.3,
n_live_points=400,
sampling_efficiency=0.3,
n_iter_before_update=2000,
)
analyzer = pymultinest.Analyzer(
outputfiles_basename=basename,
n_params=model.n_params,
)
lnZ = analyzer.get_stats()['global evidence']
print(':: Evidence Z:', lnZ / np.log(10))
return analyzer
def multinest(self,*args,**kwargs):
import pymultinest
#res = self.fit(False,True)
#f = open("calls.txt","w+")
self.freeParameters = self.modelManager.getFreeParameters()
def prior(cube, ndim, nparams):
for i,p in enumerate(self.freeParameters.values()):
cube[i] = p.prior.multinestCall(cube[i])
pass
pass
def loglike(cube, ndim, nparams):
logL = self.minusLogLike(cube)*(-1)
if(numpy.isnan(logL)):
logL = -1e10
#f.write(" ".join(map(lambda x:"%s" %x,cube[:ndim])))
#f.write(" %s\n" % logL)
return logL
pass
if('verbose' not in kwargs):
kwargs['verbose'] = True
if('resume' not in kwargs):
kwargs['resume'] = False
if('outputfiles_basename' not in kwargs):
kwargs['outputfiles_basename'] = '_1_'
pass
kwargs['log_zero'] = -1e9
pymultinest.run(loglike, prior, len(self.freeParameters), *args, **kwargs)
print("done")
#Collect the samples
analyzer = pymultinest.Analyzer(n_params=len(self.freeParameters),outputfiles_basename=kwargs['outputfiles_basename'])
eqw = analyzer.get_equal_weighted_posterior()
self.samples = eqw[:,:-1]
self.posteriors = eqw[:,-1]
def run(gp):
pymultinest.run(myloglike, myprior, gp.ndim, n_params = gp.ndim+1,
n_clustering_params = gp.nrho, # gp.ndim, or separate modes on the rho parameters only: gp.nrho
wrapped_params = [ gp.pops, gp.nipol, gp.nrho],
importance_nested_sampling = False, # INS enabled
multimodal = False, # separate modes
const_efficiency_mode = True, # use const sampling efficiency
n_live_points = gp.nlive,
evidence_tolerance = 0.0, # 0 to keep algorithm working indefinitely
sampling_efficiency = 0.05, # 0.05, MultiNest README for >30 params
n_iter_before_update = 2, # output after this many iterations
null_log_evidence = -1e100,
max_modes = gp.nlive, # preallocation of modes: max=number of live points
mode_tolerance = -1.e100, # mode tolerance in the case where no special value exists: highly negative
outputfiles_basename = gp.files.outdir,
seed = -1, verbose = True,
resume = gp.restart,
context = 0, write_output = True,
log_zero = -1e500, # points with log likelihood<log_zero will be neglected
max_iter = 0, # set to 0 for never reaching max_iter (no stopping criterium based on number of iterations)
init_MPI = False, dump_callback = None)
def run():
import gl_file as gfile
if gp.getnewdata:
gfile.bin_data()
gfile.get_data()
## number of dimensions
n_dims = gp.nepol + gp.pops*gp.nepol + gp.pops*gp.nbeta #rho, (nu, beta)_i
parameters = stringlist(gp.pops, gp.nepol)
# show live progress
# progress = pymultinest.ProgressPlotter(n_params = n_dims)
# progress.start()
# threading.Timer(2, show, [gp.files.outdir+'/phys_live.points.pdf']).start()
# print(str(len(gp.files.outdir))+': len of gp.files.outdir')
pymultinest.run(myloglike, myprior,
n_dims, n_params = n_dims, # None beforehands
n_clustering_params = gp.nepol, # separate modes on the rho parameters only
wrapped_params = None, # do not wrap-around parameters
importance_nested_sampling = True, # INS enabled
multimodal = True, # separate modes
const_efficiency_mode = True, # use const sampling efficiency
n_live_points = gp.nlive,
evidence_tolerance = 0.0, # set to 0 to keep algorithm working indefinitely
sampling_efficiency = 0.80,
n_iter_before_update = gp.nlive, # output after this many iterations
null_log_evidence = -1, # separate modes if logevidence > this param.
max_modes = gp.nlive, # preallocation of modes: maximum = number of live points
mode_tolerance = -1.,
outputfiles_basename = gp.files.outdir,
seed = -1,
verbose = True,
resume = False,
context = 0,
write_output = True,
log_zero = -1e6,
max_iter = 10000000,
init_MPI = True,
dump_callback = None)
def run(self, clean_up=None, **kwargs):
if clean_up is None:
if self.run_dir is None:
clean_up = True
else:
clean_up = False
if self.run_dir is None:
run_dir = tempfile.mkdtemp()
else:
run_dir = self.run_dir
basename = self.prepare_fit_directory(run_dir, self.prefix)
start_time = time.time()
logger.info('Starting fit in {0} with prefix {1}'.format(run_dir, self.prefix))
pymultinest.run(self.likelihood.multinest_evaluate, self.priors.prior_transform,
self.n_params,
outputfiles_basename='{0}_'.format(basename),
**kwargs)
logger.info("Fit finished - took {0:.2f} s"
.format(time.time() - start_time))
fitted_parameter_names = [item for item in self.likelihood.param_names
if not self.likelihood.fixed[item]]
self.result = MultiNestResult.from_multinest_basename(
basename, fitted_parameter_names)
if clean_up:
logger.info("Cleaning up - deleting {0}".format(run_dir))
shutil.rmtree(run_dir)
else:
logger.info("Multinest files can be found in {0}".format(run_dir))
self.likelihood.parameters[~self.likelihood.fixed_mask()] = (
self.result.median.values)
return self.result
def run(self, clean_up=None, **kwargs):
if clean_up is None:
if self.run_dir is None:
clean_up = True
else:
clean_up = False
if self.run_dir is None:
run_dir = tempfile.mkdtemp()
else:
run_dir = self.run_dir
basename = self.prepare_fit_directory(run_dir, self.prefix)
start_time = time.time()
logger.info('Starting fit in {0} with prefix {1}'.format(run_dir, self.prefix))
pymultinest.run(self.likelihood, self.priors.prior_transform,
self.n_params,
outputfiles_basename='{0}_'.format(basename),
**kwargs)
logger.info("Fit finished - took {0:.2f} s"
.format(time.time() - start_time))
self.result = MultinestResult.from_multinest_basename(
basename, self.likelihood.param_names)
if clean_up == True:
logger.info("Cleaning up - deleting {0}".format(run_dir))
shutil.rmtree(run_dir)
return self.result
try: loglike = kg_loglikelihood(cube,my_observable,my_error,a) return loglike except: return -np.inf parameters = ['Separation','Position Angle','Contrast'] n_params = len(parameters) resume=False eff=0.3 multi=True, max_iter= 0 ndim = n_params pymultinest.run(myloglike_kg, myprior, n_params, wrapped_params=[1], verbose=True,resume=False) thing = pymultinest.Analyzer(n_params = n_params) s = thing.get_stats() this_j = trial kseps[this_j], dkseps[this_j] = s['marginals'][0]['median'], s['marginals'][0]['sigma'] kthetas[this_j], dkthetas[this_j] = s['marginals'][1]['median'], s['marginals'][1]['sigma'] kcons[this_j], dkcons[this_j] = s['marginals'][2]['median'], s['marginals'][2]['sigma'] stuff = thing.get_best_fit() best_params = stuff['parameters'] model_vises = np.sqrt(pysco.binary_model(best_params,a,hdr,vis2=True)) model_kervises = np.dot(KerGain,model_vises-1.)
def main():
# MAIN -- TRADES + pyMultiNest
# ---
# initialize logger
logger = logging.getLogger("Main_log")
logger.setLevel(logging.DEBUG)
formatter = logging.Formatter("%(asctime)s - %(message)s")
# READ COMMAND LINE ARGUMENTS
cli = get_args()
# STARTING TIME
start = time.time()
# RENAME
working_path = cli.full_path
#nthreads=cli.nthreads
log_file = os.path.join(working_path, '%s_log.txt' %(os.path.dirname(cli.sub_folder)))
flog = logging.FileHandler(log_file, 'w')
flog.setLevel(logging.DEBUG)
flog.setFormatter(formatter)
logger.addHandler(flog)
# log screen
slog = logging.StreamHandler()
slog.setLevel(logging.DEBUG)
slog.setFormatter(formatter)
logger.addHandler(slog)
fitting_priors, fitting_priors_type = read_priors(os.path.join(working_path, 'fitting_priors.dat'))
derived_priors, derived_priors_type = read_priors(os.path.join(working_path, 'derived_priors.dat'))
n_der_priors = len(derived_priors)
# INITIALISE TRADES WITH SUBROUTINE WITHIN TRADES_LIB -> PARAMETER NAMES, MINMAX, INTEGRATION ARGS, READ DATA ...
pytrades_lib.pytrades.initialize_trades(working_path, cli.sub_folder, 1)
# RETRIEVE DATA AND VARIABLES FROM TRADES_LIB MODULE
fitting_parameters = pytrades_lib.pytrades.fitting_parameters # INITIAL PARAMETER SET (NEEDED ONLY TO HAVE THE PROPER ARRAY/VECTOR)
parameters_minmax = pytrades_lib.pytrades.parameters_minmax # PARAMETER BOUNDARIES
delta_parameters = np.abs(parameters_minmax[:,1] - parameters_minmax[:,0]) # DELTA BETWEEN MAX AND MIN OF BOUNDARIES
n_bodies = pytrades_lib.pytrades.n_bodies # NUMBER OF TOTAL BODIES OF THE SYSTEM
n_planets = n_bodies - 1 # NUMBER OF PLANETS IN THE SYSTEM
ndata = pytrades_lib.pytrades.ndata # TOTAL NUMBER OF DATA AVAILABLE
npar = pytrades_lib.pytrades.npar # NUMBER OF TOTAL PARAMATERS ~n_planets X 6
nfit = pytrades_lib.pytrades.nfit # NUMBER OF PARAMETERS TO FIT
nfree = pytrades_lib.pytrades.nfree # NUMBER OF FREE PARAMETERS (ie nrvset)
dof = pytrades_lib.pytrades.dof # NUMBER OF DEGREES OF FREEDOM = NDATA - NFIT
global inv_dof
#inv_dof = np.float64(1.0 / dof)
inv_dof = pytrades_lib.pytrades.inv_dof
str_len = pytrades_lib.pytrades.str_len
temp_names = pytrades_lib.pytrades.get_parameter_names(nfit,str_len)
trades_names = anc.convert_fortran_charray2python_strararray(temp_names)
parameter_names = anc.trades_names_to_emcee(trades_names)
# RADIAL VELOCITIES SET
n_rv = pytrades_lib.pytrades.nrv
n_set_rv = pytrades_lib.pytrades.nrvset
# TRANSITS SET
n_t0 = pytrades_lib.pytrades.nt0
n_t0_sum = pytrades_lib.pytrades.ntts
n_set_t0 = 0
for i in range(0, n_bodies-1):
if (n_t0[i] > 0): n_set_t0 += 1
# compute global constant for the loglhd
global ln_err_const
#try:
#e_RVo = np.asarray(pytrades_lib.pytrades.ervobs[:], dtype=np.float64) # fortran variable RV in python will be rv!!!
#except:
#e_RVo = np.asarray([0.], dtype=np.float64)
#try:
#e_T0o = np.asarray(pytrades_lib.pytrades.et0obs[:,:], dtype=np.float64).reshape((-1))
#except:
#e_T0o = np.asarray([0.], dtype=np.float64)
#ln_err_const = anc.compute_ln_err_const(ndata, dof, e_RVo, e_T0o, cli.ln_flag)
ln_err_const = pytrades_lib.pytrades.ln_err_const
# READ THE NAMES OF THE PARAMETERS FROM THE TRADES_LIB AND CONVERT IT TO PYTHON STRINGS
#reshaped_names = pytrades_lib.pytrades.parameter_names.reshape((10,nfit), order='F').T
#parameter_names = [''.join(reshaped_names[i,:]).strip() for i in range(0,nfit)]
# INITIALISE SCRIPT FOLDER/LOG FILE
working_folder, run_log, of_run = init_folder(working_path, cli.sub_folder)
logger.info('')
logger.info('==================== ')
logger.info('pyTRADES-pyMultiNest')
logger.info('==================== ')
logger.info('')
logger.info('WORKING PATH = %s' %(working_path))
logger.info('dof = ndata(%d) - nfit(%d) = %d' %(ndata, nfit, dof))
logger.info('Total N_RV = %d for %d set(s)' %(n_rv, n_set_rv))
logger.info('Total N_T0 = %d for %d out of %d planet(s)' %(n_t0_sum, n_set_t0, n_planets))
logger.info('%s = %.7f' %('log constant error = ', ln_err_const))
#of_run.write(' dof = ndata(%d) - nfit(%d) = %d\n' %(ndata, nfit, dof))
#of_run.write(' Total N_RV = %d for %d set(s)\n' %(n_rv, n_set_rv))
#of_run.write(' Total N_T0 = %d for %d out of %d planet(s)\n' %(n_t0_sum, n_set_t0, n_planets))
#of_run.write(' %s = %.7f\n' %('log constant error = ', ln_err_const))
# save parameter_names and boundaries to be read by a script
trades_hdf5 = h5py.File(os.path.join(working_folder, 'system_summary.hdf5'), 'w')
trades_hdf5.create_dataset('parameter_names', data=parameter_names, dtype='S10')
trades_hdf5.create_dataset('parameters_minmax', data=parameters_minmax, dtype=np.float64)
trades_hdf5.create_dataset('ln_err_const', data=np.asarray([ln_err_const], dtype=np.float64), dtype=np.float64)
trades_hdf5.close()
# MULTINEST HERE
#output_mnest = os.path.join(working_folder, 'trades_mnest_')
#output_mnest = os.path.join(cli.sub_folder, 'trades_mnest_')
output_mnest = 'trades_mnest_'
os.chdir(working_folder)
log_zero_value = -0.5*1.e8
seed_value = convert_to_int(cli.seed_value)
#seed_value = 392919
#n_pop = nfit+int(nfit*0.5)
n_pop = convert_to_int(cli.n_pop)
if( n_pop < nfit): n_pop = nfit*10
n_update = 5 # by argument
resume_flag = cli.resume_flag
multi_node_flag = True
mpi_onoff = False
logger.info('pyMultiNest parameters:')
logger.info('folder = %s' %(output_mnest))
logger.info('seed_value = %d , n_pop = %d , n_update = %d' %(seed_value, n_pop, n_update))
logger.info('resume_flag = %r , multi_node_flag = %r, mpi_onoff = %r' %(resume_flag, multi_node_flag, mpi_onoff))
#of_run.write(' pyMultiNest parameters:\n')
#of_run.write(' folder = %s\n seed_value = %d , n_pop = %d , n_update = %d\n resume_flag = %r , multi_node_flag = %r, mpi_onoff = %r\n' %(output_mnest, seed_value, n_pop, n_update, resume_flag, multi_node_flag, mpi_onoff))
#
# RESCALE PARAMETERS FUNCTION NEEDED BY LNLIKE
#
def trades_rescale(fitting_parameters, ndim, nparams):
for i in range(0,ndim):
fitting_parameters[i] = parameters_minmax[i,0] + fitting_parameters[i]*delta_parameters[i]
return fitting_parameters
# LNPRIOR TO BE ADDED TO LOGLHD
# it can use all the variables defined before this point!
def lnprior(fitting_parameters, ndim):
lnprior_value = 0.
i_der = 0
for i in range(0, ndim):
#print i,parameter_names[i], fitting_priors_type[i]
ln_temp = 0.
# calculate the LogLikelihood<->prior of fitting parameter
if(fitting_priors_type[i][1].lower() == 'g'):
ln_temp = -0.5*(((fitting_parameters[i]-fitting_priors[i][0])/fitting_priors[i][1])**2)
lnprior_value = lnprior_value + ln_temp
#print '%18.12f %18.12f (%18.12f) => ln = %18.12f' %(fitting_parameters[i], fitting_priors[i][0], fitting_priors[i][1], ln_temp)
# calculate the LogLikelihood<->prior of derived parameter
if('mA' in parameter_names[i]):
ln_temp = 0.
ecc = np.sqrt(fitting_parameters[i-2]**2 + fitting_parameters[i-1]**2)
if (ecc <= 0.):
ecc = np.finfo(float).eps
elif (ecc > 1.):
ecc = 1.- np.finfo(float).eps
# ecc prior
if(derived_priors_type[i_der][1].lower() == 'g'):
ln_temp = -0.5*(((ecc-derived_priors[i_der][0])/derived_priors[i_der][1])**2)
lnprior_value = lnprior_value + ln_temp
#print derived_priors_type[i_der]
#print '%18.12f %18.12f (%18.12f) => ln = %18.12f' %(ecc, derived_priors[i_der][0], derived_priors[i_der][1], ln_temp)
# phi prior
if(derived_priors_type[i_der+1][1].lower() == 'g'):
if(ecc <= np.finfo(float).eps):
argp = 90.
else:
argp = ((np.arctan2(fitting_parameters[i-1], fitting_parameters[i-2])*180./np.pi)+360.)%360.
phi = (argp + fitting_parameters[i] + 360.)%360.
ln_temp = 0.
ln_temp = -0.5*(((phi-derived_priors[i_der+1][0])/derived_priors[i_der+1][1])**2)
lnprior_value = lnprior_value + ln_temp
#print derived_priors_type[i_der+1]
#print '%18.12f (argp[%18.12f]+mA[%18.12f]) %18.12f (%18.12f) => ln = %18.12f' %(phi, argp, fitting_parameters[i], derived_priors[i_der+1][0], derived_priors[i_der+1][1], ln_temp)
i_der = i_der + 2
return lnprior_value
# LNLIKEHOOD FUNCTION NEEDED BY MULTINEST
def lnlike(fitting_parameters, ndim, nparams):
loglhd = 0.
check = 1
trades_parameters = np.asarray([fitting_parameters[i] for i in range(0,ndim)], dtype=np.float64)
loglhd, check = pytrades_lib.pytrades.fortran_loglikelihood(trades_parameters)
if ( check == 0 ):
loglhd = -0.5e10
else:
lnprior_value = lnprior(fitting_parameters, ndim)
#lnprior_value = 0.0
loglhd = loglhd + ln_err_const + lnprior_value # ln_err_const (global variable) & lnprior_value
return loglhd
# run MultiNest
pymultinest.run(LogLikelihood = lnlike, Prior = trades_rescale, n_dims = nfit, n_params = nfit, outputfiles_basename=output_mnest, multimodal = multi_node_flag, log_zero=log_zero_value, seed = seed_value, n_live_points = n_pop, n_iter_before_update = n_update, resume = resume_flag, verbose = True, init_MPI = mpi_onoff)
elapsed = time.time() - start
elapsed_d, elapsed_h, elapsed_m, elapsed_s = computation_time(elapsed)
logger.info('')
logger.info('pyTRADES: pyMultiNest FINISHED in %2d day %02d hour %02d min %.2f sec - bye bye' %(int(elapsed_d), int(elapsed_h), int(elapsed_m), elapsed_s))
logger.info('')
#of_run.write(' pyTRADES: pyMultiNest FINISHED in %2d day %02d hour %02d min %.2f sec - bye bye\n' %(int(elapsed_d), int(elapsed_h), int(elapsed_m), elapsed_s))
#of_run.close()
pytrades_lib.pytrades.deallocate_variables()
return
def nested_run(id=None, otherids=(), prior = None, parameters = None,
sampling_efficiency = 0.3, evidence_tolerance = 0.5,
n_live_points = 400, outputfiles_basename = 'chains/', **kwargs):
"""
Run the Bayesian analysis with specified parameters+transformations.
:param id: See the sherpa documentation of calc_stat.
:param otherids: See the sherpa documentation of calc_stat.
:param prior: prior function created with create_prior_function.
:param parameters: List of parameters to analyse.
:param outputfiles_basename: prefix for output filenames.
If prior is None, uniform priors are used on the passed parameters.
If parameters is also None, all thawed parameters are used.
The remainder are multinest arguments (see PyMultiNest and MultiNest documentation!)
n_live_points: 400 are often enough
For quick results, use sampling_efficiency = 0.8, n_live_points = 50,
evidence_tolerance = 5.
The real results must be estimated with sampling_efficiency = 0.3,
otherwise it is not reliable.
"""
fit = ui._session._get_fit(id=id, otherids=otherids)[1]
if not isinstance(fit.stat, (Cash, CStat)):
raise RuntimeError("Fit statistic must be cash or cstat, not %s" %
fit.stat.name)
if parameters is None:
parameters = fit.model.thawedpars
def log_likelihood(cube, ndim, nparams):
try:
for i, p in enumerate(parameters):
assert not isnan(cube[i]), 'ERROR: parameter %d (%s) to be set to %f' % (i, p.fullname, cube[i])
p.val = cube[i]
#print "%s: %f" % (p.fullname,p.val),
l = -0.5*fit.calc_stat()
#print "%.1f" % l
return l
except Exception as e:
print 'Exception in log_likelihood function: ', e
for i, p in enumerate(parameters):
print ' Parameter %10s: %f --> %f [%f..%f]' % (p.fullname, p.val, cube[i], p.min, p.max)
import sys
sys.exit(-127)
return -1e300
if prior is None:
prior = create_prior_function(id=id, otherids=otherids, parameters = parameters)
n_params = len(parameters)
pymultinest.run(log_likelihood, prior, n_params,
sampling_efficiency = sampling_efficiency, n_live_points = n_live_points,
outputfiles_basename = outputfiles_basename, evidence_tolerance=evidence_tolerance,
**kwargs)
import json
m = ui._session._get_model(id)
paramnames = map(lambda x: x.fullname, parameters)
json.dump(paramnames, file('%sparams.json' % outputfiles_basename, 'w'), indent=2)
def myloglike_kg(cube,ndim,n_params): try: loglike = kg_loglikelihood(cube,my_observable,my_error,a) return loglike except: return -np.inf parameters = ['Semi-major axis','Eccentricity','Position Angle', 'Thickness','Contrast'] n_params = len(parameters) resume=False eff=0.3 multi=True, max_iter= 10000 ndim = n_params pymultinest.run(myloglike_kg, myprior, n_params,wrapped_params=[2], verbose=False,resume=False,max_iter=max_iter) thing = pymultinest.Analyzer(n_params = n_params) try: s = thing.get_stats() ksemis[trial], dksemis[trial] = s['marginals'][0]['median']/4., s['marginals'][0]['sigma']/4. keccs[trial], dkeccs[trial] = s['marginals'][1]['median'], s['marginals'][1]['sigma'] kthetas[trial], dkthetas[trial] = s['marginals'][2]['median'], s['marginals'][2]['sigma'] kthicks[trial], dkthicks[trial] = s['marginals'][3]['median'], s['marginals'][3]['sigma'] kcons[trial], dkcons[trial] = s['marginals'][4]['median'], s['marginals'][4]['sigma'] stuff = thing.get_best_fit() best_params = stuff['parameters'] print 'Best params (kg):', best_params
def multinest_run(n_live_points=1000,
target = 'ob110022', saveto=multiNest_saveto):
def priors(cube, ndim, nparams):
return
def likelihood(cube, ndim, nparams):
lnlikePhot = 0.0
lnlike_nonPhot = 0.0
parnames = Photpars['parnames']
# Photometric params
if MultiDimPriors == True:
params, lnlikePhot = random_photpar_multidim(cube[0],
Photpars['MultiDimPrior_Bins'],
Photpars['MultiDimPrior'][:,0],
Photpars['MultiDimPrior'][:,1],
Photpars['MultiDimPrior'][:, -nPhotpars:])
for i in range(nPhotpars): cube[i] = params[i]
else:
for i in range(nPhotpars):
param, ln_prob_param = random_photpar(cube[i],
Photpars[parnames[i]][:,0],
Photpars[parnames[i]][:,1],
Photpars[parnames[i]][:,2])
cube[i]=param
lnlikePhot += ln_prob_param
idx = nPhotpars
# x Position at t0:
thetaS0x, ln_prob_thetaS0x = random_thetaS0x(cube[idx])
cube[idx] = thetaS0x
idx += 1
lnlike_nonPhot += ln_prob_thetaS0x
# y Position at t0:
thetaS0y, ln_prob_thetaS0y = random_thetaS0y(cube[idx])
cube[idx] = thetaS0y
idx += 1
lnlike_nonPhot += ln_prob_thetaS0y
# Source proper motion (x dimension)
muSx, ln_prob_muSx = random_muSx(cube[idx])
cube[idx] = muSx
idx += 1
lnlike_nonPhot += ln_prob_muSx
# Source proper motion (y dimension)
muSy, ln_prob_muSy = random_muSy(cube[idx])
cube[idx] = muSy
idx += 1
lnlike_nonPhot += ln_prob_muSy
# Source-lens relative proper motion (x dimension)
muRelx, ln_prob_muRelx = random_muRelx(cube[idx])
cube[idx] = muRelx
idx += 1
lnlike_nonPhot += ln_prob_muRelx
# Source-lens relative proper motion (y dimension)
muRely, ln_prob_muRely = random_muRely(cube[idx])
cube[idx] = muRely
idx += 1
lnlike_nonPhot += ln_prob_muRely
t0 = cube[0]
beta = cube[1]
tE = cube[2]
piEN = cube[3]
piEE= cube[4]
#Create astrometric model of source
thetaS_model, thetaE_amp, M, shift, thetaS_nolens = MCMC_LensModel.LensModel_Trial1(tobs, t0, tE,
[thetaS0x, thetaS0y],
[muSx,muSy],
[muRelx, muRely],
beta,
[piEN, piEE])
cube[11] = thetaE_amp
cube[12] = M
thetaSx_model = thetaS_model[:,0]
thetaSy_model = thetaS_model[:,1]
lnlike = lnlikePhot + lnlike_nonPhot + \
MCMC_LensModel.lnLikelihood(thetaSx_model, thetaSx_data, xerr_data) + \
MCMC_LensModel.lnLikelihood(thetaSy_model, thetaSy_data, yerr_data)
# print "Log Likelihood: ", lnlike
return lnlike
## num_dims = 11
## num_params = 13
num_dims= 11
num_params= 13 #cube will have this many dimensions
ev_tol=0.3
samp_eff=0.8
#Create param file
_run = open(saveto + runcode + '_params.run', 'w')
_run.write('Num Dimensions: %d\n' % num_dims)
_run.write('Num Params: %d\n' % num_params)
_run.write('Evidence Tolerance: %.1f\n' % ev_tol)
_run.write('Sampling Efficiency: %.1f\n' % samp_eff)
_run.write('Num Live Points: %d\n' % n_live_points)
_run.close()
startdir = os.getcwd()
os.chdir(saveto)
pymultinest.run(likelihood, priors, num_dims, n_params=num_params,
outputfiles_basename= runcode + '_',
verbose=True, resume=False, evidence_tolerance=ev_tol,
sampling_efficiency=samp_eff, n_live_points=n_live_points,
multimodal=True, n_clustering_params=num_dims,
importance_nested_sampling=False)
os.chdir(startdir)
# Testing
# lnlike = multinest_run.likelihood([0.01,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.95], 12, 12)
# print lnlike
return
def nest(kpo,paramlimits=[20.,250.,0.,360.,1.0001,10],ndim=3,resume=False,eff=0.3,multi=True,
max_iter=0,bispec=False):
'''Default implementation of a MultiNest fitting routine for kernel
phase data. Requires a kernel phase kpo object, parameter limits and
sensible keyword arguments for the multinest parameters.
This function does very naughty things creating functions inside this
function because PyMultiNest is very picky about how you pass it
data.
Optional parameter eff tunes sampling efficiency, and multi toggles multimodal
nested sampling on and off. Turning off multimodal sampling results in a speed
boost of ~ 20-30%.
'''
import pymultinest # importing here so you don't have to unless you use nest!
# make sure you're using the right number of parameters
nbands = kpo.kpd.shape[0]
# if 'WFC3' in kpo.hdr['tel']:
# bands = str(round(1e9*kpo.hdr['filter'],3))
# parameters = ['Separation','Position Angle','Contrast at ' + bands + ' nm']
# print bands
# print parameters
# else:
if np.size(kpo.hdr) == 1:
bands = str(round(1e6*kpo.hdr['filter'],3))
parameters = ['Separation','Position Angle','Contrast at ' + bands + ' um']
else:
bands = [str(round(1e6*hd['filter'],3)) for hd in kpo.hdr]
parameters = ['Separation','Position Angle'] + ['Contrast at ' + band + ' um' for band in bands]
n_params = len(parameters)
ndim = n_params
def myprior(cube, ndim, n_params,paramlimits=paramlimits):
cube[0] = (paramlimits[1] - paramlimits[0])*cube[0]+paramlimits[0]
cube[1] = (paramlimits[3] - paramlimits[2])*cube[1]+paramlimits[2]
for j in range(2,ndim):
cube[j] = (paramlimits[5] - paramlimits[4])*cube[j]+paramlimits[4]
if bispec:
print 'Using a bispectral analysis'
def myloglike(cube,ndim,n_params):
loglike = bispec_loglikelihood(cube,kpo)
return loglike
else:
print 'Modelling kernel phases with nested sampling'
def myloglike(cube,ndim,n_params):
loglike = kp_loglikelihood(cube,kpo)
return loglike
tic = time.time() # start timing
#---------------------------------
# now run MultiNest!
#---------------------------------
pymultinest.run(myloglike, myprior, n_params, wrapped_params=[1], resume=resume,
verbose=True, sampling_efficiency=eff, multimodal=multi,
n_iter_before_update=1000,max_iter=max_iter)
# let's analyse the results
a = pymultinest.Analyzer(n_params = n_params)
s = a.get_stats()
toc = time.time()
if toc-tic < 60.:
print 'Time elapsed =',toc-tic,'s'
else:
print 'Time elapsed =',(toc-tic)/60.,'mins'
null = -0.5*np.sum(((kpo.kpd)/kpo.kpe)**2)
# json.dump(s, file('%s.json' % a.outputfiles_basename, 'w'), indent=2)
print
print "-" * 30, 'ANALYSIS', "-" * 30
print "Global Evidence:\n\t%.15e +- %.15e" % ( s['global evidence']-null, s['global evidence error'] )
params = s['marginals']
bestsep = params[0]['median']
seperr = params[0]['sigma']
if 'Hale' in kpo.hdr['tel']: params[1]['median'] += 220.0 + kpo.hdr['orient']
elif 'HST' in kpo.hdr['tel']: params[1]['median'] -= kpo.hdr['orient']
else: params[1]['median'] += 0.0
params[1]['median'] = np.mod(params[1]['median'],360.)
bestth = params[1]['median']
therr = params[1]['sigma']
print 'Separation: %.3f pm %.2f' % (bestsep,seperr)
print 'Position angle: %.3f pm %.2f' %(bestth,therr)
if kpo.nsets == 1:
bestcon = params[2]['median']
conerr = params[2]['sigma']
print 'Contrast at',bands,'um: %.3f pm %.3f' % (bestcon,conerr)
else:
for j, band in enumerate(bands):
bestcon = params[j+2]['median']
conerr = params[j+2]['sigma']
print 'Contrast at',band,'um: %.3f pm %.3f' % (bestcon,conerr)
return params
def sample_multinest(self, n_live_points, chain_name="chains/fit-", quiet=False, **kwargs):
"""
Sample the posterior with MULTINEST nested sampling (Feroz & Hobson)
:param: n_live_points: number of MULTINEST livepoints
:param: chain_names: where to stor the multinest incremental output
:param: quiet: Whether or not to should results
:param: **kwargs (pyMULTINEST kwords)
:return: MCMC samples
"""
assert has_pymultinest, "You don't have pymultinest installed, so you cannot run the Multinest sampler"
self._update_free_parameters()
n_dim = len(self._free_parameters.keys())
# MULTINEST has a convergence criteria and therefore, there is no way
# to determine progress
sampling_procedure = sample_without_progress
# MULTINEST uses a different call signiture for
# sampling so we construct callbakcs
loglike, multinest_prior = self._construct_multinest_posterior()
# We need to check if the MCMC
# chains will have a place on
# the disk to write and if not,
# create one
mcmc_chains_out_dir = ""
tmp = chain_name.split('/')
for s in tmp[:-1]:
mcmc_chains_out_dir += s + '/'
if using_mpi:
# if we are running in parallel and this is not the
# first engine, then we want to wait and let everything finish
if rank != 0:
# let these guys take a break
time.sleep(1)
else:
# create mcmc chains directory only on first engine
if not os.path.exists(mcmc_chains_out_dir):
os.makedirs(mcmc_chains_out_dir)
else:
if not os.path.exists(mcmc_chains_out_dir):
os.makedirs(mcmc_chains_out_dir)
print("\nSampling\n")
print("MULTINEST has its own convergence criteria... you will have to wait blindly for it to finish")
print("If INS is enabled, one can monitor the likelihood in the terminal for completion information")
# Multinest must be run parallel via an external method
# see the demo in the examples folder!!
if threeML_config['parallel']['use-parallel']:
raise RuntimeError("If you want to run multinest in parallell you need to use an ad-hoc method")
else:
sampler = pymultinest.run(loglike,
multinest_prior,
n_dim,
n_dim,
outputfiles_basename=chain_name,
n_live_points=n_live_points,
**kwargs)
# Use PyMULTINEST analyzer to gather parameter info
process_fit = False
if using_mpi:
# if we are running in parallel and this is not the
# first engine, then we want to wait and let everything finish
if rank !=0:
# let these guys take a break
time.sleep(5)
# these engines do not need to read
process_fit = False
else:
# wait for a moment to allow it all to turn off
time.sleep(5)
process_fit = True
else:
process_fit = True
if process_fit:
multinest_analyzer = pymultinest.analyse.Analyzer(n_params=n_dim,
outputfiles_basename=chain_name)
# Get the log. likelihood values from the chain
self._log_like_values = multinest_analyzer.get_equal_weighted_posterior()[:, -1]
self._sampler = sampler
self._raw_samples = multinest_analyzer.get_equal_weighted_posterior()[:, :-1]
# now get the log probability
self._log_probability_values = self._log_like_values + np.array([self._log_prior(samples) for samples in self._raw_samples])
self._build_samples_dictionary()
self._marginal_likelihood = multinest_analyzer.get_stats()['global evidence'] / np.log(10.)
self._build_results()
# Display results
if not quiet:
self._results.display()
# now get the marginal likelihood
return self.samples
print "\n You are searching for the following parameters: {0}\n".format(parameters)
n_params = len(parameters)
print "\n The total number of parameters is {0}\n".format(n_params)
#####################
# Now, we sample.....
#####################
print "\n Now, we sample... \n"
if args.eccSearch==True:
dirextension = 'eccSearch'
else:
dirextension = 'circSearch'
master_path = os.getcwd()
os.chdir(args.datapath)
pymultinest.run(lnprob, my_prior_mnest, n_params,
importance_nested_sampling = False,
resume = False, verbose = True, n_live_points=1000,
outputfiles_basename=u'chains_{0}/{0}_'.format(dirextension),
sampling_efficiency='parameter')
import pymultinest as pmn
import pandas as pd
import triangle
from mn_test import *
pmn.run(llhood_w_const, prior_transform_const, 3, verbose=True, n_live_point = 150)
data = pd.read_csv('chains/1-post_equal_weights.dat', names=['H0', 'om', 'w', 'lhood'], delim_whitespace=True, header=False)
triangle.corner(data.values[:,:-1], labels = data.columns[:-1], smooth=1)
def Explore(self):
'''
This member function invokes multinest.
The data must be loaded and the likihood set
'''
print bcolors.WARNING+'___________________________________________________'+bcolors.ENDC
print bcolors.WARNING+'|'+bcolors.OKBLUE+' _____ _____ _____ _ _ '+bcolors.WARNING+'|'+bcolors.ENDC
print bcolors.WARNING+'|'+bcolors.OKBLUE+'| | ___ | __| ___ ___ ___ | __||_|| |_ '+bcolors.WARNING+'|'+bcolors.ENDC
print bcolors.WARNING+'|'+bcolors.OKBLUE+'| | | || ||__ || . || -_|| _|| __|| || _|'+bcolors.WARNING+'|'+bcolors.ENDC
print bcolors.WARNING+'|'+bcolors.OKBLUE+'|_|_|_||_|_||_____|| _||___||___||__| |_||_| '+bcolors.WARNING+'|'+bcolors.ENDC
print bcolors.WARNING+'|'+bcolors.OKBLUE+' |_| '+bcolors.WARNING+'|'+bcolors.ENDC
print bcolors.WARNING+'| |'+bcolors.ENDC
print bcolors.WARNING+'| |'+bcolors.ENDC
print bcolors.WARNING+'| |'+bcolors.ENDC
print bcolors.WARNING+'| '+bcolors.OKGREEN+'-J. Michael Burgess'+bcolors.WARNING+' |'+bcolors.ENDC
print bcolors.WARNING+'| |'+bcolors.ENDC
print bcolors.WARNING+'|_________________________________________________|'+bcolors.ENDC
#print
#print "%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%"
#print
#print "License Info:"
#print "\t Don't read this.\n\t Do whatever the hell you want with this software"
#print
#print "%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%"
if not self._dataLoaded: #Make sure to have loaded data
print
print bcolors.FAIL+ "YOU HAVE NOT LOADED ANY DATA!!"+ bcolors.ENDC
print
return
if not self._saveFileSet: #Warn that no savefile is set
print
print bcolors.WARNING+"Save file not set!!! Fit params not saved!"+ bcolors.ENDC
print
outfilesDir = ""
tmp = self.basename.split('/')
for s in tmp[:-1]:
outfilesDir+=s+'/'
self.outfilesDir = outfilesDir
if not os.path.exists(outfilesDir): os.makedirs(outfilesDir)
# we want to see some output while it is running
if not self.silent:
print "SILENT"
progress = pymultinest.ProgressPlotter(n_params = self.n_params); progress.start()
threading.Timer(2, show, [self.basename+"phys_live.points.pdf"]).start() # delayed opening
startTime = time.time()
self._PreFitInfo()
# run MultiNest
pymultinest.run(self.likelihood, self.prior, self.n_params, importance_nested_sampling = self.importance_nested_sampling, resume = self.resume, verbose = self.verbose, sampling_efficiency = self.sampling_efficiency, n_live_points = self.n_live_points,outputfiles_basename=self.basename, init_MPI=False, dump_callback=self.callback,write_output=self.write)
# ok, done. Stop our progress watcher
if not self.silent:
progress.stop()
print
print bcolors.OKGREEN +"Finished sampling in %.2f seconds"%(time.time()-startTime) + bcolors.ENDC
print
if self._saveFileSet:
self._WriteFit()
def run(catalogfile, vel_err, mag_err, N_gauss, outdir, rotate=True):
"""
PyMultiNest run to determine cluster membership, using PM catalog and
applying vel_err and mag_err cuts. Output is put in newly-created outdir
directory (must be a string).
Parameters:
catalogflie --> String containing the name of a FITS catalog.
vel_err --> The maximum allowed velocity error for stars to be included.
mag_err --> The maximum allowed magnitude error for stars to be included.
N_gauss --> number bivariate gaussian, where N_gauss <= 4
outdir --> The output directory name.
Keywords:
rotate = 1 --> rotate star velocities into RA/DEC format, as opposed
to X,Y
"""
# Load data for full field, extract velocities (already converted to mas)
d = loadData(catalogfile, vel_err, mag_err, rotate=rotate)
star_Vx = d["fit_vx"]
star_Vy = d["fit_vy"]
star_Sigx = d["fit_vxe"]
star_Sigy = d["fit_vye"]
N_stars = len(d)
def print_param(pname, val, logp, headerFirst=False):
rowHead = "{0:6s} "
colHead = " val_{0} ( logp_{0} )"
colVal = "{0:6.3f} ({1:9.2e})"
if headerFirst:
outhdr = " ".join([colHead.format(k) for k in range(N_gauss)])
print rowHead.format("") + outhdr
outstr = " ".join([colVal.format(val[k], logp[k]) for k in range(N_gauss)])
print rowHead.format(pname) + outstr
return
def priors(cube, ndim, nparams):
return
def likelihood(cube, ndim, nparams):
"""
Define the likelihood function (from Clarkson+12, Hosek+15)
"""
# start the timer
t0 = time.time()
####################
# Set up model params
####################
# Number of parameters per Gaussian:
N_per_gauss = 6
# Make arrays for the paramters of each Gaussian
pi = np.arange(N_gauss, dtype=float)
vx = np.arange(N_gauss, dtype=float)
vy = np.arange(N_gauss, dtype=float)
sigA = np.arange(N_gauss, dtype=float)
sigB = np.arange(N_gauss, dtype=float)
theta = np.arange(N_gauss, dtype=float)
# Make arrays for the prior probability of each paramter
logp_pi = np.arange(N_gauss, dtype=float)
logp_vx = np.arange(N_gauss, dtype=float)
logp_vy = np.arange(N_gauss, dtype=float)
logp_sigA = np.arange(N_gauss, dtype=float)
logp_sigB = np.arange(N_gauss, dtype=float)
logp_theta = np.arange(N_gauss, dtype=float)
# Set the fraction of stars in each Gaussian
for kk in range(N_gauss):
pi[kk], logp_pi[kk] = random_pi(cube[kk * N_per_gauss + 0])
# Make sure all the sum(pi) = 1.
pi /= pi.sum()
# Sort the field pi values such that they are always ranked from
# smallest to largest.
sidx = pi[1:].argsort()
pi[1:] = pi[1:][sidx]
logp_pi[1:] = logp_pi[1:][sidx]
# Re-set the cube values. Note this is AFTER sorting.
for kk in range(N_gauss):
cube[kk * N_per_gauss + 0] = pi[kk]
# Set the other Gaussian parameters.
for kk in range(N_gauss):
# Treat the cluster gaussian (the first, most compact one)
# with a special prior function.
if kk == 0:
rand_vx = random_clust_vx
rand_vy = random_clust_vy
rand_sigA = random_clust_sigA
rand_sigB = random_clust_sigB
rand_theta = random_clust_theta
else:
rand_vx = random_v
rand_vy = random_v
rand_sigA = random_sig
rand_sigB = random_sig
rand_theta = random_theta
# Velocity centr
vx[kk], logp_vx[kk] = rand_vx(cube[kk * N_per_gauss + 1])
cube[kk * N_per_gauss + 1] = vx[kk]
vy[kk], logp_vy[kk] = rand_vy(cube[kk * N_per_gauss + 2])
cube[kk * N_per_gauss + 2] = vy[kk]
# Major axis
sigA[kk], logp_sigA[kk] = rand_sigA(cube[kk * N_per_gauss + 3])
cube[kk * N_per_gauss + 3] = sigA[kk]
# Minor axis
sigB[kk], logp_sigB[kk] = rand_sigB(cube[kk * N_per_gauss + 4])
cube[kk * N_per_gauss + 4] = sigB[kk]
# Angle of major axis (in radians)
theta[kk], logp_theta[kk] = rand_theta(cube[kk * N_per_gauss + 5])
cube[kk * N_per_gauss + 5] = theta[kk]
# Only want to consider gaussians where Sig A > Sig B
if sigB[kk] > sigA[kk]:
# print '#######################'
# print '#######################'
# print '#######################'
# print '#######################'
return -np.Inf
# Check that all our prior probabilities are valid, otherwise abort
# before expensive calculation.
if (
(logp_pi[kk] == -np.inf)
or (logp_vx[kk] == -np.inf)
or (logp_vy[kk] == -np.inf)
or (logp_sigA[kk] == -np.inf)
or (logp_sigB[kk] == -np.inf)
or (logp_theta[kk] == -np.inf)
):
return -np.Inf
################################
# Calculating likelihood function
# Likelihood =
# \Sum(i=0 -> N_stars) \Sum(k=0 -> N_gauss)
# \pi_k * (2 \pi |\Sigma_k,i|)^{-1/2} *
# exp[ -1/2 * (\mu_i - \mu_k)^T \sigma_k,i (\mu_i - \mu_k) ]
################################
# Keep track of the probability for each star, each gaussian
# component. We will add over components and multiply over stars.
prob_gauss = np.zeros((N_gauss, N_stars), dtype=float)
# L_{i,k} Loop through the different gaussian components.
for kk in range(N_gauss):
# N_stars long array
prob_gauss[kk, :] = prob_ellipse(
star_Vx, star_Vy, star_Sigx, star_Sigy, pi[kk], vx[kk], vy[kk], sigA[kk], sigB[kk], theta[kk]
)
# For each star, the total likelihood is the sum
# of each component (before log).
L_star = prob_gauss.sum(axis=0) # This array should be N-stars long
logL_star = np.log10(L_star)
# Final likelihood
logL = logL_star.sum()
logL_tmp = logL
# Add in log(prior probabilities) as well
for kk in range(N_gauss):
logL += logp_pi[kk]
logL += logp_vx[kk]
logL += logp_vy[kk]
logL += logp_sigA[kk]
logL += logp_sigB[kk]
logL += logp_theta[kk]
# Some printing
print "*** logL = {0:9.2e} w/priors = {1:9.2e}".format(logL_tmp, logL)
print_param("pi", pi, logp_pi, headerFirst=True)
print_param("vx", vx, logp_vx)
print_param("vy", vy, logp_vy)
print_param("sigA", sigA, logp_sigA)
print_param("sigB", sigB, logp_sigB)
print_param("theta", theta, logp_theta)
t1 = time.time()
total = t1 - t0
print "TIME SPENT: " + str(total)
# pdb.set_trace()
return logL
#########################################
# End Likelihoods
# Begin running multinest
#########################################
# Make new directory to hold output
fileUtil.mkdir(outdir)
outroot = outdir + "/mnest_"
num_dims = 2 * 3 * N_gauss
num_params = num_dims
ev_tol = 0.3
samp_eff = 0.8
n_live_points = 300
# Create param file
_run = open(outroot + "params.run", "w")
_run.write("Catalog: %s\n" % catalogfile)
_run.write("Vel Err Cut: %.2f\n" % vel_err)
_run.write("Mag Err Cut: %.2f\n" % mag_err)
_run.write("Rotate: %s\n" % str(rotate))
_run.write("Num Gauss: %d\n" % N_gauss)
_run.write("Num Dimensions: %d\n" % num_dims)
_run.write("Num Params: %d\n" % num_params)
_run.write("Evidence Tolerance: %.1f\n" % ev_tol)
_run.write("Sampling Efficiency: %.1f\n" % samp_eff)
_run.write("Num Clustering Params: %d\n" % num_dims)
_run.write("Num Live Points: %d\n" % n_live_points)
_run.close()
# Run multinest
pymultinest.run(
likelihood,
priors,
num_dims,
n_params=num_params,
outputfiles_basename=outroot,
verbose=True,
resume=False,
evidence_tolerance=ev_tol,
sampling_efficiency=samp_eff,
n_live_points=n_live_points,
multimodal=True,
n_clustering_params=num_dims,
importance_nested_sampling=False,
)
return
mpirun -np 4 python run_PyPolyChord.py
on Mac:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$PWD/lib
export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libgfortran.so.3
export LD_PRELOAD=/opt/local/lib/openmpi//lib/libmpi.so:$LD_PRELOAD
mpirun -np 4 python run_PyPolyChord.py
'''
# number of dimensions our problem has
parameters = mc.variable_list
n_params = mc.ndim
pymultinest.run(mc.multinest_call, mc.multinest_priors, n_params, resume=True,
verbose=True, n_live_points=1000, outputfiles_basename=dir_output + 'chains/2-')
def dump():
progress = pymultinest.ProgressPlotter(n_params = n_params, outputfiles_basename=dir_output+'chains/2-'); progress.start()
threading.Timer(2, show, [dir_output+"chains/2-phys_live.points.pdf"]).start() # delayed opening
# run MultiNest
pymultinest.run(mc.multinest_call, mc.multinest_priors, n_params, importance_nested_sampling = False, resume = True,
verbose = True, sampling_efficiency = 'model', n_live_points = 1000, outputfiles_basename=dir_output+'chains/2-')
# ok, done. Stop our progress watcher
progress.stop()
# lets analyse the results
a = pymultinest.Analyzer(n_params = n_params, outputfiles_basename=dir_output+'chains/2-')
s = a.get_stats()
def dump():
progress = pymultinest.ProgressPlotter(n_params = n_params, outputfiles_basename=dir_output+'chains/2-'); progress.start()
threading.Timer(2, show, [dir_output+"chains/2-phys_live.points.pdf"]).start() # delayed opening
# run MultiNest
pymultinest.run(mc.multinest_call, mc.multinest_priors, n_params, importance_nested_sampling = False, resume = True,
verbose = True, sampling_efficiency = 'model', n_live_points = 1000, outputfiles_basename=dir_output+'chains/2-')
# ok, done. Stop our progress watcher
progress.stop()
# lets analyse the results
a = pymultinest.Analyzer(n_params = n_params, outputfiles_basename=dir_output+'chains/2-')
s = a.get_stats()
# store name of parameters, always useful
with open('%sparams.json' % a.outputfiles_basename, 'w') as f:
json.dump(parameters, f, indent=2)
# store derived stats
with open('%sstats.json' % a.outputfiles_basename, mode='w') as f:
json.dump(s, f, indent=2)
print()
print("-" * 30, 'ANALYSIS', "-" * 30)
print("Global Evidence:\n\t%.15e +- %.15e" % ( s['nested sampling global log-evidence'], s['nested sampling global log-evidence error'] ))
import matplotlib.pyplot as plt
plt.clf()
# run MultiNest
#pymultinest.run(mc.pymultinest_call, mc.pymultinest_priors, mc.ndim, outputfiles_basename=dir_output, resume = False, verbose = True)
#json.dump(parameters, open(dir_output+'params.json', 'w')) # save parameter names
# Here we will plot all the marginals and whatnot, just to show off
# You may configure the format of the output here, or in matplotlibrc
# All pymultinest does is filling in the data of the plot.
# Copy and edit this file, and play with it.
p = pymultinest.PlotMarginalModes(a)
plt.figure(figsize=(5 * n_params, 5 * n_params))
# plt.subplots_adjust(wspace=0, hspace=0)
for i in range(n_params):
plt.subplot(n_params, n_params, n_params * i + i + 1)
p.plot_marginal(i, with_ellipses=True, with_points=False, grid_points=50)
plt.ylabel("Probability")
plt.xlabel(parameters[i])
for j in range(i):
plt.subplot(n_params, n_params, n_params * j + i + 1)
# plt.subplots_adjust(left=0, bottom=0, right=0, top=0, wspace=0, hspace=0)
p.plot_conditional(i, j, with_ellipses=False, with_points=True, grid_points=30)
plt.xlabel(parameters[i])
plt.ylabel(parameters[j])
plt.savefig(dir_output+"chains/marginals_multinest.pdf") # , bbox_inches='tight')
show(dir_output+"chains/marginals_multinest.pdf")
for i in range(n_params):
outfile = '%s-mode-marginal-%d.pdf' % (a.outputfiles_basename, i)
p.plot_modes_marginal(i, with_ellipses=True, with_points=False)
plt.ylabel("Probability")
plt.xlabel(parameters[i])
plt.savefig(outfile, format='pdf', bbox_inches='tight')
plt.close()
outfile = '%s-mode-marginal-cumulative-%d.pdf' % (a.outputfiles_basename, i)
p.plot_modes_marginal(i, cumulative=True, with_ellipses=True, with_points=False)
plt.ylabel("Cumulative probability")
plt.xlabel(parameters[i])
plt.savefig(outfile, format='pdf', bbox_inches='tight')
plt.close()
print("Take a look at the pdf files in chains/")
def nested_run(transformations, prior_function = None, sampling_efficiency = 0.3,
n_live_points = 400, evidence_tolerance = 0.5,
outputfiles_basename = 'chains/', verbose=True, **kwargs):
"""
Run the Bayesian analysis with specified parameters+transformations.
If prior is None, uniform priors are used on the passed parameters.
If parameters is also None, all thawed parameters are used.
:param transformations: Parameter transformation definitions
:param prior_function: set only if you want to specify a custom, non-separable prior
:param outputfiles_basename: prefix for output filenames.
The remainder are multinest arguments (see PyMultiNest and MultiNest documentation!)
The remainder are multinest arguments (see PyMultiNest and MultiNest documentation!)
n_live_points: 400 are often enough
For quick results, use sampling_efficiency = 0.8, n_live_points = 50,
evidence_tolerance = 5.
The real results must be estimated with sampling_efficiency = 0.3
and without using const_efficiency_mode, otherwise it is not reliable.
"""
# for convenience
if outputfiles_basename.endswith('/'):
if not os.path.exists(outputfiles_basename):
os.mkdir(outputfiles_basename)
if prior_function is None:
prior_function = create_prior_function(transformations)
oldchatter = Xset.chatter, Xset.logChatter
Xset.chatter, Xset.logChatter = 0, 0
def log_likelihood(cube, ndim, nparams):
try:
set_parameters(transformations=transformations, values=cube)
l = -0.5*Fit.statistic
#print "like = %.1f" % l
return l
except Exception as e:
print 'Exception in log_likelihood function: ', e
import sys
sys.exit(-127)
return -1e300
# run multinest
if Fit.statMethod.lower() not in ['cstat', 'cash']:
raise RuntimeError('ERROR: not using cash (Poisson likelihood) for Poisson data! set Fit.statMethod to cash before analysing (currently: %s)!' % Fit.statMethod)
n_params = len(transformations)
pymultinest.run(log_likelihood, prior_function, n_params,
sampling_efficiency = sampling_efficiency, n_live_points = n_live_points,
outputfiles_basename = outputfiles_basename,
verbose=verbose, **kwargs)
paramnames = [str(t['name']) for t in transformations]
json.dump(paramnames, file('%sparams.json' % outputfiles_basename, 'w'), indent=4)
# store as chain too, and try to load it for error computations
analyzer = pymultinest.Analyzer(n_params = len(transformations),
outputfiles_basename = outputfiles_basename)
posterior = analyzer.get_equal_weighted_posterior()
chainfilename = '%schain.fits' % outputfiles_basename
store_chain(chainfilename, transformations, posterior)
xspec.AllChains.clear()
Xset.chatter, Xset.logChatter = oldchatter
xspec.AllChains += chainfilename
# set current parameters to best fit
set_best_fit(analyzer, transformations)
Xset.chatter, Xset.logChatter = oldchatter
return analyzer
chi = 1.
#print "cube", [cube[i] for i in range(ndim)], cube
for i in range(ndim):
chi *= math.cos(cube[i] / 2.)
#print "returning", math.pow(2. + chi, 5)
return math.pow(2. + chi, 5)
# number of dimensions our problem has
parameters = ["x", "y"]
n_params = len(parameters)
# we want to see some output while it is running
progress = pymultinest.ProgressPlotter(n_params = n_params, outputfiles_basename='chains/2-'); progress.start()
threading.Timer(2, show, ["chains/2-phys_live.points.pdf"]).start() # delayed opening
# run MultiNest
pymultinest.run(myloglike, myprior, n_params, importance_nested_sampling = False, resume = True, verbose = True, sampling_efficiency = 'model', n_live_points = 1000, outputfiles_basename='chains/2-')
# ok, done. Stop our progress watcher
progress.stop()
# lets analyse the results
a = pymultinest.Analyzer(n_params = n_params, outputfiles_basename='chains/2-')
s = a.get_stats()
import json
# store name of parameters, always useful
with open('%sparams.json' % a.outputfiles_basename, 'w') as f:
json.dump(parameters, f, indent=2)
# store derived stats
with open('%sstats.json' % a.outputfiles_basename, mode='w') as f:
json.dump(s, f, indent=2)
print()
