def optimise(self, problem, rundir=None):
if rundir is not None:
self.dump(filename=op.join(rundir, 'optimiser.yaml'))
history = ModelHistory(problem,
nchains=self.nchains,
path=rundir, mode='w')
chains = self.chains(problem, history)
niter = self.niterations
isbad_mask = None
self._tlog_last = 0
for iiter in range(niter):
iphase, phase, iiter_phase = self.get_sampler_phase(iiter)
self.log_progress(problem, iiter, niter, phase, iiter_phase)
sample = phase.get_sample(problem, iiter_phase, chains)
sample.iphase = iphase
if isbad_mask is not None and num.any(isbad_mask):
isok_mask = num.logical_not(isbad_mask)
else:
isok_mask = None
misfits = problem.misfits(sample.model, mask=isok_mask)
bootstrap_misfits = problem.combine_misfits(
misfits,
extra_weights=self.get_bootstrap_weights(problem),
extra_residuals=self.get_bootstrap_residuals(problem))
isbad_mask_new = num.isnan(misfits[:, 0])
if isbad_mask is not None and num.any(
isbad_mask != isbad_mask_new):
errmess = [
'problem %s: inconsistency in data availability'
' at iteration %i' %
(problem.name, iiter)]
for target, isbad_new, isbad in zip(
problem.targets, isbad_mask_new, isbad_mask):
if isbad_new != isbad:
errmess.append(' %s, %s -> %s' % (
target.string_id(), isbad, isbad_new))
raise BadProblem('\n'.join(errmess))
isbad_mask = isbad_mask_new
if num.all(isbad_mask):
raise BadProblem(
'Problem %s: all target misfit values are NaN.'
% problem.name)
history.append(
sample.model, misfits,
bootstrap_misfits,
sample.pack_context())
def get_history(self, subset=None):
if subset not in self._histories:
self._histories[subset] = \
ModelHistory(
self.get_problem(),
nchains=self.get_optimiser().nchains,
path=meta.xjoin(self.get_rundir_path(), subset))
return self._histories[subset]