def _do_backward_pass(self, framelogprob):
n_samples, n_components = framelogprob.shape
bwdlattice = np.zeros((n_samples, n_components))
_hmmc._backward(n_samples, n_components,
log_mask_zero(self.startprob_),
log_mask_zero(self.transmat_), framelogprob,
bwdlattice)
return bwdlattice
def _do_backward_pass(log_startprob, log_transmat, framelogprob):
n_samples, n_components = framelogprob.shape
bwdlattice = np.zeros((n_samples, n_components))
_hmmc._backward(
n_samples, n_components, log_startprob, log_transmat, framelogprob,
bwdlattice
)
return bwdlattice
def _do_backward_pass(self, framelogprob):
# Based on hmmlearn's _BaseHMM
safe_startmat = self.startprob_ + np.finfo(float).eps
safe_transmat = self.transmat_ + np.finfo(float).eps
n_samples, n_components = framelogprob.shape
bwdlattice = np.zeros((n_samples, n_components))
_hmmc._backward(n_samples, n_components, np.log(safe_startmat),
np.log(safe_transmat), framelogprob, bwdlattice)
return bwdlattice
def _do_backward_pass(self, framelogprob):
# Based on hmmlearn's _BaseHMM
safe_startmat = self.startprob_ + np.finfo(float).eps
safe_transmat = self.transmat_ + np.finfo(float).eps
n_samples, n_components = framelogprob.shape
bwdlattice = np.zeros((n_samples, n_components))
_hmmc._backward(n_samples, n_components,
np.log(safe_startmat),
np.log(safe_transmat),
framelogprob, bwdlattice)
return bwdlattice
int(n_components), \
torch.from_numpy(np.log(startprob)), \
torch.from_numpy(np.log(transmat)), \
torch.from_numpy(data), \
torch.from_numpy(mask))
print("torch batch forward: \n {}\n".format(torch_fwdlattice))
print("torch batch logprob: {}\n".format(torch_logprob))
# 3. backward comparison
hmmlearn_bwdlattice = []
for idx, framelogprob in enumerate(data):
bwdlattice = np.zeros((mask[idx].sum(), n_components))
_hmmc._backward(int(mask[idx].sum()), \
int(n_components), \
np.log(startprob), \
np.log(transmat), \
framelogprob[mask[idx]>0], \
bwdlattice)
hmmlearn_bwdlattice.append(bwdlattice)
print("hmmlearn backward lattice: \n{}".format(hmmlearn_bwdlattice))
# 3.2 batch backward in PyTorch
torch_bwdlattice = _backward(int(data.shape[1]), \
int(n_components), \
torch.from_numpy(np.log(startprob)), \
torch.from_numpy(np.log(transmat)), \
torch.from_numpy(data), \
torch.from_numpy(mask))
print("torch batch backward: \n{}".format(torch_bwdlattice))
def _do_backward_pass(self, framelogprob):
n_observations, n_components = framelogprob.shape
bwdlattice = np.zeros((n_observations, n_components))
_hmmc._backward(n_observations, n_components, self._log_startprob,
self._log_transmat, framelogprob, bwdlattice)
return bwdlattice
