def hmm(self, nstates):
"""Estimates a hidden Markov state model as described in [1]_
Returns
-------
hmsm : :class:`MaximumLikelihoodHMSM`
References
----------
.. [1] F. Noe, H. Wu, J.-H. Prinz and N. Plattner:
Projected and hidden Markov models for calculating kinetics and metastable states of complex molecules
J. Chem. Phys. 139, 184114 (2013)
"""
self._check_is_estimated()
# run estimate
from pyemma.msm.estimators.maximum_likelihood_hmsm import MaximumLikelihoodHMSM
estimator = MaximumLikelihoodHMSM(lag=self.lagtime,
nstates=nstates,
msm_init=self,
reversible=self.is_reversible,
connectivity=self.connectivity,
observe_active=True,
dt_traj=self.dt_traj)
estimator.estimate(self.discrete_trajectories_full)
return estimator.model
def submodel(self, states=None, obs=None, mincount_connectivity='1/n'):
# call submodel on MaximumLikelihoodHMSM
_MaximumLikelihoodHMSM.submodel(self, states=states, obs=obs, mincount_connectivity=mincount_connectivity)
# if samples set, also reduce them
if hasattr(self, 'samples') and self.samples is not None:
subsamples = [sample.submodel(states=self.active_set, obs=self.observable_set)
for sample in self.samples]
self.update_model_params(samples=subsamples)
# return
return self
def hmm(self, nhidden):
"""Estimates a hidden Markov state model as described in [1]_
Parameters
----------
nhidden : int
number of hidden (metastable) states
Returns
-------
hmsm : :class:`MaximumLikelihoodHMSM`
References
----------
.. [1] F. Noe, H. Wu, J.-H. Prinz and N. Plattner:
Projected and hidden Markov models for calculating kinetics and metastable states of complex molecules
J. Chem. Phys. 139, 184114 (2013)
"""
self._check_is_estimated()
# check if the time-scale separation is OK
# if hmm.nstates = msm.nstates there is no problem. Otherwise, check spectral gap
if self.nstates > nhidden:
timescale_ratios = self.timescales()[:-1] / self.timescales()[1:]
if timescale_ratios[nhidden - 2] < 1.5:
self.logger.warning(
'Requested coarse-grained model with {nhidden} metastable states at lag={lag}.'
' The ratio of relaxation timescales between'
' {nhidden} and {nhidden_1} states is only {ratio}'
' while we recommend at least 1.5.'
' It is possible that the resulting HMM is inaccurate. Handle with caution.'
.format(
lag=self.lag,
nhidden=nhidden,
nhidden_1=nhidden + 1,
ratio=timescale_ratios[nhidden - 2],
))
# run HMM estimate
from pyemma.msm.estimators.maximum_likelihood_hmsm import MaximumLikelihoodHMSM
estimator = MaximumLikelihoodHMSM(lag=self.lagtime,
nstates=nhidden,
msm_init=self,
reversible=self.is_reversible,
dt_traj=self.dt_traj)
estimator.estimate(self.discrete_trajectories_full)
return estimator.model