Version numbers here correspond to git tags in the repository history.
Some additional features that we found useful for implementing the experiments for our recent paper on debiasing gradient estimators for deep latent variable models. Most significantly, this release adds the ability to train using different objectives, including IWAE and reweighted wake-sleep, and different gradient estimators, including the doubly reparameterized and sticking the landing gradient estimators. This release also contains a variety of small fixes and usability improvements.
Initial release of CIF version of paper (an earlier version released for the LGF version of the paper is available at https://github.com/jrmcornish/lgf)