def init(self,
log=False,
inv=False,
defaultprior=True,
priorshape=False,
priors=(),
originalscale=True,
scale=1,
sigma_m=False,
p_M=.5,
simarf=None,
p_M_arf=.5,
sigma_arf=0.1):
# nsubiters is missing from init() to indicate that nubiters=1 for
# full bayes
if isinstance(simarf, PCA1DAdd):
self.simarf = simarf
else:
self.simarf = ARFSIMFactory()(simarf)
if not isinstance(self.simarf, PCA1DAdd):
raise TypeError("Simulation ARF must be PCA for FullBayes" +
" not %s" % type(self.simarf).__name__)
self.accept_arfs = [0]
self.p_M_arf = p_M_arf
self.rrsig = sigma_arf
return MetropolisMH.init(self, log, inv, defaultprior, priorshape,
priors, originalscale, scale, sigma_m, p_M)
def init(self, log=False, inv=False, defaultprior=True, priorshape=False,
priors=(), originalscale=True, scale=1, sigma_m=False, p_M=.5,
simarf=None, p_M_arf=.5, sigma_arf=0.1):
# nsubiters is missing from init() to indicate that nubiters=1 for
# full bayes
if isinstance(simarf, (PCA1DAdd,)):
self.simarf = simarf
else:
self.simarf = ARFSIMFactory()(simarf)
if not isinstance(self.simarf, (PCA1DAdd,)):
raise TypeError("Simulation ARF must be PCA for FullBayes" +
" not %s" % type(self.simarf).__name__)
self.accept_arfs = [0]
self.p_M_arf = p_M_arf
self.rrsig = sigma_arf
return MetropolisMH.init(self, log, inv, defaultprior, priorshape,
priors, originalscale, scale, sigma_m, p_M)
class FullBayes(PragBayes):
def __init__(self, fcn, sigma, mu, dof, fit, *args):
PragBayes.__init__(self, fcn, sigma, mu, dof, fit, *args)
self.arf_dicts = [{'current' : arf.specresp.copy(), 'old_rr' : None }
for arf in self.arfs]
def init(self, log=False, inv=False, defaultprior=True, priorshape=False,
priors=(), originalscale=True, scale=1, sigma_m=False, p_M=.5,
simarf=None, p_M_arf=.5, sigma_arf=0.1):
# nsubiters is missing from init() to indicate that nubiters=1 for
# full bayes
if isinstance(simarf, (PCA1DAdd,)):
self.simarf = simarf
else:
self.simarf = ARFSIMFactory()(simarf)
if not isinstance(self.simarf, (PCA1DAdd,)):
raise TypeError("Simulation ARF must be PCA for FullBayes" +
" not %s" % type(self.simarf).__name__)
self.accept_arfs = [0]
self.p_M_arf = p_M_arf
self.rrsig = sigma_arf
return MetropolisMH.init(self, log, inv, defaultprior, priorshape,
priors, originalscale, scale, sigma_m, p_M)
def _update_arf(self, arf, specresp, current_params, current_stat, arf_dict):
u = np.random.uniform(0,1,1)
if u > self.p_M_arf:
ncomp = self.simarf.ncomp
old_rr = arf_dict['old_rr']
if old_rr is None:
old_rr = np.random.standard_normal(ncomp)
# Assume the ARF is accepted by default
# Update the ARFs with new deviates
arf.specresp = self.simarf.add_deviations(specresp, old_rr, self.rrsig)
new_rr = self.simarf.rrout
stat_temp = self.calc_fit_stat(self._mu)
mu0 = np.repeat(0,ncomp)
sig0 = np.diag(np.repeat(1,ncomp))
accept_pr = dmvnorm(new_rr,mu0,sig0)-dmvnorm(old_rr,mu0,sig0)
accept_pr += stat_temp - current_stat
accept_pr = np.exp( accept_pr )
uu = np.random.uniform(0,1,1)
if accept_pr > uu:
arf_dict['old_rr'] = new_rr
arf_dict['current'] = arf.specresp.copy()
self.accept_arf()
# When ARF is updated, set scale to None to signal a fit
self._sigma = None
else:
# Restore to previous ARF
arf.specresp = arf_dict['current']
self.reject_arf()
else:
# Assume the ARF is accepted by default
# Update the ARFs with new deviates
arf.specresp = self.simarf.add_deviations(specresp)
stat_temp = self.calc_fit_stat(self._mu)
accept_pr=0
accept_pr += stat_temp - current_stat
accept_pr = np.exp( accept_pr )
uu = np.random.uniform(0,1,1)
if accept_pr > uu:
arf_dict['current'] = arf.specresp.copy()
self.accept_arf()
# When ARF is updated, set scale to None to signal a fit
self._sigma = None
else:
# Restore to previous ARF
arf.specresp = arf_dict['current']
self.reject_arf()
def accept_arf(self):
# Accept the updated ARF deviates
self.accept_arfs.append(self.accept_arfs[-1] + 1)
def reject_arf(self):
# Reject
self.accept_arfs.append(self.accept_arfs[-1])
def perturb_arf(self, current_params, current_stat):
if self.simarf is not None:
# add deviations starting with original ARF for each iter
for specresp, arf, arf_dict in zip(self.backup_arfs, self.arfs, self.arf_dicts):
self._update_arf(arf, specresp, current_params, current_stat, arf_dict)
class FullBayes(PragBayes):
def __init__(self, fcn, sigma, mu, dof, fit, *args):
PragBayes.__init__(self, fcn, sigma, mu, dof, fit, *args)
self.arf_dicts = [{
'current': arf.specresp.copy(),
'old_rr': None
} for arf in self.arfs]
def init(self,
log=False,
inv=False,
defaultprior=True,
priorshape=False,
priors=(),
originalscale=True,
scale=1,
sigma_m=False,
p_M=.5,
simarf=None,
p_M_arf=.5,
sigma_arf=0.1):
# nsubiters is missing from init() to indicate that nubiters=1 for
# full bayes
if isinstance(simarf, PCA1DAdd):
self.simarf = simarf
else:
self.simarf = ARFSIMFactory()(simarf)
if not isinstance(self.simarf, PCA1DAdd):
raise TypeError("Simulation ARF must be PCA for FullBayes" +
" not %s" % type(self.simarf).__name__)
self.accept_arfs = [0]
self.p_M_arf = p_M_arf
self.rrsig = sigma_arf
return MetropolisMH.init(self, log, inv, defaultprior, priorshape,
priors, originalscale, scale, sigma_m, p_M)
def _update_arf(self, arf, specresp, current_params, current_stat,
arf_dict):
u = np.random.uniform(0, 1, 1)
if u > self.p_M_arf:
ncomp = self.simarf.ncomp
old_rr = arf_dict['old_rr']
if old_rr is None:
old_rr = np.random.standard_normal(ncomp)
# Assume the ARF is accepted by default
# Update the ARFs with new deviates
arf.specresp = self.simarf.add_deviations(specresp, old_rr,
self.rrsig)
new_rr = self.simarf.rrout
stat_temp = self.calc_fit_stat(self._mu)
mu0 = np.repeat(0, ncomp)
sig0 = np.diag(np.repeat(1, ncomp))
accept_pr = dmvnorm(new_rr, mu0, sig0) - dmvnorm(old_rr, mu0, sig0)
accept_pr += stat_temp - current_stat
accept_pr = np.exp(accept_pr)
uu = np.random.uniform(0, 1, 1)
if accept_pr > uu:
arf_dict['old_rr'] = new_rr
arf_dict['current'] = arf.specresp.copy()
self.accept_arf()
# When ARF is updated, set scale to None to signal a fit
self._sigma = None
else:
# Restore to previous ARF
arf.specresp = arf_dict['current']
self.reject_arf()
else:
# Assume the ARF is accepted by default
# Update the ARFs with new deviates
arf.specresp = self.simarf.add_deviations(specresp)
stat_temp = self.calc_fit_stat(self._mu)
accept_pr = 0
accept_pr += stat_temp - current_stat
accept_pr = np.exp(accept_pr)
uu = np.random.uniform(0, 1, 1)
if accept_pr > uu:
arf_dict['current'] = arf.specresp.copy()
self.accept_arf()
# When ARF is updated, set scale to None to signal a fit
self._sigma = None
else:
# Restore to previous ARF
arf.specresp = arf_dict['current']
self.reject_arf()
def accept_arf(self):
# Accept the updated ARF deviates
self.accept_arfs.append(self.accept_arfs[-1] + 1)
def reject_arf(self):
# Reject
self.accept_arfs.append(self.accept_arfs[-1])
def perturb_arf(self, current_params, current_stat):
if self.simarf is not None:
# add deviations starting with original ARF for each iter
for specresp, arf, arf_dict in zip(self.backup_arfs, self.arfs,
self.arf_dicts):
self._update_arf(arf, specresp, current_params, current_stat,
arf_dict)
