eb8680
eb8680
NumPyro now has several excellent introductory [examples](http://num.pyro.ai/en/stable/) with no direct counterparts in Pyro. Porting one of these to Pyro would be a great way for someone to simultaneously learn more...
Motivated by off-target Sphinx build errors in #3006. Can't remember if there was a reason we weren't doing this, feel free to close if there was one...
This PR makes a few small changes to Pyro's `EnumMessenger`, `MarkovMessenger`, `EnumAllocator` and `DimAllocator` which allow the use of existing Pyro models and primitives with `pyro.contrib.funsor` (especially @ordabayevy's new inference...
This PR is an attempt at making a couple small changes to the Pyro and `PyroModule` API to make `PyroModule` more compatible with vanilla PyTorch programming idioms. The API changes...
The error-checking logic in `check_site_shape` in `TraceEnum_ELBO` incorrectly raises an error when performing guide-side enumeration in a model/guide pair where the factorization differs between model and guide such that the...
Pyro's HMC and NUTS implementations are feature-complete and well-tested, but they are quite slow in models like the one in our [Bayesian regression tutorial](http://pyro.ai/examples/bayesian_regression.html) that operate on small tensors for...
This issue proposes adding an implementation of [nonparametric Hamiltonian Monte Carlo](https://arxiv.org/abs/2106.10238), a modified version of HMC that provably correctly handles models with stochastic control flow, unlike vanilla HMC. @fzaiser (one...
Now that PyTorch supports tensor subtyping and function overloading with `__torch_function__`, should we add `__array_function__` and `__torch_function__` methods to `funsor.terms.Funsor` to allow evaluation of (some) PyTorch/Numpy code on Funsors? Here...
This PR fixes a longstanding alpha-conversion edge case by guaranteeing that newly bound variables are always alpha-renamed, which is currently not the case due to the way alpha-conversion interacts with...