Is your feature request related to a problem? Please describe.
I was wondering why it's only possible to compute the likelihoods post-hoc. Computing likelihood themselves requires integrating the system again, that can obviously be quite time consuming. Many applications require a reweighting of the samples, so it would be convenient to obtain the likelihoods with the samples.
Describe the solution you'd like
The likelihood computation is based on the change of variable theorem, where the log determinants of the Jacobian get summed up over the timesteps. So it should be possible to simply compute them along the way with the sampling process.
I am thinking of an extra flag to the sampling method to additionally obtain the likelhoods or not.
Is your feature request related to a problem? Please describe.
I was wondering why it's only possible to compute the likelihoods post-hoc. Computing likelihood themselves requires integrating the system again, that can obviously be quite time consuming. Many applications require a reweighting of the samples, so it would be convenient to obtain the likelihoods with the samples.
Describe the solution you'd like
The likelihood computation is based on the change of variable theorem, where the log determinants of the Jacobian get summed up over the timesteps. So it should be possible to simply compute them along the way with the sampling process.
I am thinking of an extra flag to the sampling method to additionally obtain the likelhoods or not.