The implementation of transition model is slightly different from that described in the paper.

[Haiyao-Nero]

Hi there, there is a place I'm confused about. In the paper, the prior $p\theta^{(z)} = p\theta(x{t},y{t},z{t-1},a)$. but in the transition model from the code, the $y{t}$ is missing, when compute the prior $p_\theta^{(z)}$.

[ssnl]

Ah good catch! The experimental results in main paper are performed with the variant without $z$ component, so thankfully they are still valid. I've fixed it now on master. Thanks for pointing it out! The paper's appendix (only place w/ experiments with $z$) will be updated soon.