numpyro
numpyro copied to clipboard
pyro-ppl
Probabilistic programming with NumPy powered by JAX for autograd and JIT compilation to GPU/TPU/CPU.
Currently, we skip deterministic sites while rendering the model. It would be nice to support plotting deterministic nodes (e.g. to make the rendered model clearer when using reparam handlers). This...
Addresses SVI doc accessibility requests/suggestions from #1309
Some time series wrapper as discussed in #1361 . Could appreciate any discussions and inputs as the effort is very nascent. A few things this model tries to accomplish: 1....
I noticed that we already have [a few time series forecasting models](https://num.pyro.ai/en/stable/tutorials/time_series_forecasting.html) as part of the examples. For these models we mostly write the models from ground up. On the...
The [GP example](https://num.pyro.ai/en/latest/examples/gp.html) directly inverts the covariance matrix. ```python K_xx_inv = jnp.linalg.inv(k_XX) K = k_pp - jnp.matmul(k_pX, jnp.matmul(K_xx_inv, jnp.transpose(k_pX))) ``` It may be better to use the Cholesky decomposition instead....
When writing custom distributions, it is often helpful to have numerically stable implementations of `log_diff_exp(a, b) := log(exp(a) - exp(b))` and particularly `log1m_exp(x) := log(1 - exp(x))`. The naive implementations...
Currently, from [the docs](https://num.pyro.ai/en/stable/distributions.html#mixture-distributions) it is not clear that the last batch dimension of `component_distribution` should match the category size. In addition, it seems the docstring is not aligned well...
I have a function which returns a posterior distribution of type `Independent` with `base_dist` being `Normal`. I want to `vmap` over the function in order to get the posteriors for...
So I was playing around with SVI API and problems from Statistical Rethinking 2ed. And one thing - extracting posterior sample data - made me quite puzzled for a while....
This PR introduces a new field `reprocess` to `sample` sites. If it is available, after: `process_msg` and `default_process_msg`, we will apply `reprocess_msg`. I added the following two handlers to illustrate...