def _twofilter_smoothing_ON(self, t, ti, info, phi, lwinfo, return_ess,
modif_forward, modif_info):
"""O(N) version of two-filter smoothing.
This method should not be called directly, see twofilter_smoothing.
"""
if modif_info is not None:
lwinfo += modif_info
Winfo = rs.exp_and_normalise(lwinfo)
I = rs.multinomial(Winfo)
if modif_forward is not None:
lw = self.wgts[t].lw + modif_forward
W = rs.exp_and_normalise(lw)
else:
W = self.wgts[t].W
J = rs.multinomial(W)
log_omega = self.fk.logpt(t + 1, self.X[t][J], info.hist.X[ti][I])
if modif_forward is not None:
log_omega -= modif_forward[J]
if modif_info is not None:
log_omega -= modif_info[I]
Om = rs.exp_and_normalise(log_omega)
est = np.average(phi(self.X[t][J], info.hist.X[ti][I]),
axis=0,
weights=Om)
if return_ess:
return (est, 1. / np.sum(Om**2))
else:
return est
def backward_sampling(self, M, linear_cost=False, return_ar=False):
"""Generate smoothing trajectories using FFBS.
FFBS (forward filtering backward smoothing) is a class of off-line
smoothing algorithms, which generate smoothing trajectories constructed
from the history of a particle filter.
Arguments
---------
M: int
number of trajectories we want to generate
linear_cost: bool
if set to True, the O(N) version is used, see below.
return_ar: bool (default=False)
if set to True, change the output, see below.
Returns
-------
paths: a list of ndarrays
paths[t][n] is component t of trajectory m.
ar: float
the overall acceptance rate of the rejection procedure
Notes
-----
1. if ``linear_cost=False``, complexity is O(TMN); i.e. O(TN^2) for M=N;
if ``linear_cost=True``, complexity is O(T(M+N)), i.e. O(TN) for M=N.
This requires that model has method `upper_bound_trans`, which
provides the log of a constant C_t such that
:math:`p_t(x_t|x_{t-1}) \leq C_t`.
2. main output is ``paths``, a list of T arrays such that
``paths[t][m]`` is component t of trajectory m.
3. if ``linear_cost=True`` and ``return_ar=True``, output is tuple
``(paths, ar)``, where ``paths`` is as above, and ``ar`` is the overall
acceptance rate (of the rejection steps that choose the ancestors);
otherwise output is simply ``paths``.
"""
idx = np.empty((self.T, M), dtype=int)
idx[-1, :] = rs.multinomial(self.wgts[-1].W, M=M)
if linear_cost:
ar = self._backward_sampling_ON(M, idx)
else:
self._backward_sampling_ON2(M, idx)
# When M=1, we want a list of states, not a list of arrays containing
# one state
if M == 1:
idx = idx.squeeze(axis=1)
paths = [self.X[t][idx[t]] for t in range(self.T)]
if linear_cost and return_ar:
return (paths, ar)
else:
return paths
def backward_sampling(self, M, linear_cost=False, return_ar=False):
"""Generate trajectories using the FFBS (forward filtering backward
sampling) algorithm.
Arguments
---------
M: int
number of trajectories we want to generate
linear_cost: bool
if set to True, the O(N) version is used, see below.
return_ar: bool (default=False)
if set to True, change the output, see below.
Returns
-------
paths: a list of ndarrays
paths[t][n] is component t of trajectory m.
ar: float
the overall acceptance rate of the rejection procedure
Notes
-----
1. if linear_cost=False, complexity is O(TMN); i.e. O(TN^2) for M=N;
if =True, complexity is O(T(M+N)), i.e. O(TN) for M=N.
This requires that model has method `upper_bound_trans(self,t)`, which
provides the log of a constant C_t such that p_t(x_t|x_{t-1})<=C_t
2. main output is *paths*, a list of T arrays such that
paths[t][m] is component t of trajectory m.
3. if linear_cost=True and return_ar=True, output is tuple (paths, ar),
where paths is as above, and ar is the overall acceptance rate
(of the rejection steps that choose the ancestors); otherwise
output is simply paths
"""
idx = np.empty((self.T, M), dtype=int)
idx[-1, :] = rs.multinomial(self.wgt[-1].W, M=M)
if linear_cost:
ar = self._backward_sampling_ON(M, idx)
else:
self._backward_sampling_ON2(M, idx)
# When M=1, we want a list of states, not a list of arrays containing
# one state
if M == 1:
idx = idx.squeeze(axis=1)
paths = [self.X[t][idx[t]] for t in range(self.T)]
if linear_cost and return_ar:
return (paths, ar)
else:
return paths
def sample(self, N=1):
"""Sample N trajectories from the posterior.
Note
----
Performs the forward step in case it has not been performed.
"""
if not self.filt:
self.forward()
paths = np.empty((len(self.filt), N), np.int)
paths[-1, :] = rs.multinomial(self.filt[-1], M=N)
log_trans = np.log(self.hmm.trans_mat)
for t, f in reversed(list(enumerate(self.filt[:-1]))):
for n in range(N):
probs = rs.exp_and_normalise(log_trans[:, paths[t + 1, n]] + np.log(f))
paths[t, n] = rs.multinomial_once(probs)
return paths