def run_minuit(self, kwargs={}):
solver = Minuit(self._core_lsq, [0.5] * len(self._activelist),
name=sorted(self._activelist))
solver.limits = (0., 1.)
solver.migrad()
best = np.array(solver.values)
solver.minos()
err = np.array(solver.errors)
names = sorted(self._activelist)
for i in range(len(names)):
low, high = self.paramrange[names[i]]
best[i] = umap(best[i], [low, high])
err[i] = umap(err[i], [low, high])
return (best, err)
def run_dynesty(self, kwargs={}):
# dynesty solver
solver = DynamicNestedSampler(self._core_likelihood, self.prior,
len(self._activelist))
solver.run_nested(**kwargs)
results = solver.results
names = sorted(self._activelist)
for i in range(len(names)):
low, high = self.paramrange[names[i]]
results.samples[:, i] = umap(results.samples[:, i], [low, high])
return results
def _core_lsq(self, cube):
if np.any(cube > 1.) or np.any(cube < 0.):
return np.nan_to_num(np.inf)
name_list = sorted(self._activelist)
for i in range(len(name_list)):
name = name_list[i]
tmp = umap(cube[i], self._paramrange[name])
if self._foreground is not None:
self._foreground.reset({name: tmp})
if self._background is not None:
self._background.reset({name: tmp})
# predict data
if self._foreground is None:
return self.lsq(self._background.bandpower())
elif self._background is None:
return self.lsq(self._foreground.bandpower())
else:
return self.lsq(self._foreground.bandpower() +
self._background.bandpower())
def run_emcee(self, kwargs={'nwalker': 100, 'nstep': 10000}):
# emcee solver
solver = emcee.EnsembleSampler(kwargs['nwalker'],
len(self._activelist),
self._core_likelihood)
state = solver.run_mcmc(
np.random.uniform(size=(kwargs['nwalker'], len(self._activelist))),
kwargs['nstep'] // 10) # burn-in
solver.reset()
state = solver.run_mcmc(state, kwargs['nstep'])
tau = int(np.mean(solver.get_autocorr_time())
) # estimate integral auto-correlation time
if (kwargs['nstep'] < 50 * tau): # if the pre-set step is not enough
solver.reset()
solver.run_mcmc(state, 100 * tau)
results = solver.get_chain(discard=10 * tau, flat=True)
names = sorted(self._activelist)
for i in range(len(names)):
low, high = self.paramrange[names[i]]
results[:, i] = umap(results[:, i], [low, high])
return results