numpyro
numpyro copied to clipboard
pyro-ppl
Probabilistic programming with NumPy powered by JAX for autograd and JIT compilation to GPU/TPU/CPU.
Currently, [summary](https://github.com/pyro-ppl/numpyro/blob/master/numpyro/diagnostics.py#L219) returns ordinary statistics mean/var/quantiles/n_eff for MCMC samples. It would be nice to extend that utility to calculate weighted statistics. This would be useful for important sampling or nested...
Since we are using python 3, we can start taking advantage of type hints wherever possible.
Initial structure and outline, with lead up to inference. Currently shows some kind of broadcasting error.
Following #548 discussion, and while we wait for discrete latent variables, it would be nice to have a Gumbel-Softmax categorical approximation as featured in Pyro. Didn't realize this was the...
Now that we have an IAF autoguide, it should be relatively straightforward to add an IAF layer to our VAE example and do some comparisons.
To get around perf issues, we have had to put in optimizations in certain parts of the codebase. Sometimes these optimizations add to code complexity and can be removed once...
A good way to get acquainted with NumPyro would be to write your model of interest and contribute it to [examples](https://github.com/pyro-ppl/numpyro/tree/master/examples) or as a [notebook](https://github.com/pyro-ppl/numpyro/tree/master/notebooks). If you are interested, please...
First draft for the MatrixNormal distribution as requested in #1178. Currently `sample` as well as `log_prob` use the close relationship between Matrix Normal and Multivariate Normal as described [here](https://en.wikipedia.org/wiki/Matrix_normal_distribution).
See preliminary discussion on this issue: https://github.com/pyro-ppl/numpyro/issues/1452.
