Closed bsugerman closed 3 years ago
Yes, you are right. The multiplication with par[0]
is currently present twice and can be removed in line 78. Thanks for noticing. It would be great if you can create a pull request for fixing the typo.
I pushed the change to a local fork and make a pull request.
Thanks for your great paper and this repo. Looks like you made a type-o in the APF update step at
https://github.com/jelfring/particle-filter-tutorial/blob/16142c850fcbbef67ebcfb7d895a617514f59357/core/particle_filters/auxiliary_particle_filter.py#L78
by multiplying the
compute_likelihood
by the previous weightpar[0]
to form the currenttmp_likelihood
estimate. As such, your temporary weight uses $w{k-1}^2$ and your final weight is (with simplified notation) $\frac{p(x|y)}{w{k-1}p(\mu|y}$.