Currently our non-notebook examples are manually included in the examples webpage via custom .rst files in tutorial/source/. As the number of examples increases, it would be better to follow NumPyro's approach and generate HTML pages automatically with sphinx_gallery. This would make examples easier t

Summary:

It'd be nice to have a builder pattern for var contexts to make them easy to construct for testing. Something that could be used like this:

MatrixXd m(3, 2); 
...
var_context vc
  = var_context::builder()
    .matrix("a", m)
    .real("f", 2.3)
    .build();

Current Version:

v2.23.0

AutoDelta is used to do MAP inference in Pyro. It would be nice to have it in NumPyro too.

References

• Pyro AutoDelta implementation
• AutoNormal implementatio

To be more in line with the rest of the regression families I think it would be a good idea to support loglinear distribution reparameterized with mean and standard deviation on the natural scale.

Here is the reparametrization (from ProbOnto, https://sites.google.com/site/probonto/download):

$P\left(x ; \boldsymbol{\mu}{N}, \boldsymbol{\sigma}{N}\right)=\frac{1}{x \sqrt{2 \pi \log \left

Hi @JavierAntoran @stratisMarkou,

First of all, thanks for making all of this code available - it's been great to look through!

Im currently spending some time trying to work through the Weight Uncertainty in Neural Networks in order to implement Bayes-by-Backprop. I was struggling to understand the difference between your implementation of `Bayes-by-Bac

There are a variety of interesting optimisations that can be performed on kernels of the form

k(x, z) = w_1 * k_1(x, z) + w_2 * k_2(x, z) + ... + w_L k_L(x, z)

A naive recursive implementation in terms of the current Sum and Scaled kernels hides opportunities for parallelism in the computation of each term, and the summation over terms.

Notable examples of kernels with th