bayes-kit

bayes-kit copied to clipboard

Make tests deterministic from seed

7

Add conftest.py and use it to set a random seed for all pytest tests. This PR makes pytest deterministic and solves the problem of tests randomly failing due to unlucky...

wrongu

Handle user-provided log densities that throw exceptions

2

This came up in a discussion with @gil2rok during #39, but applies to all of our algorithms. The user provided log_density_gradient function may throw an exception (we may even _expect_...

WardBrian

Clarifying Draw Terminology

3

Per my discussion with @WardBrian , in HMC a draw can refer to two distinct concepts: 1. The position variable `theta`, which denotes a draw from the *target distribution* 2....

gil2rok

Improve Gradient Caching in DRGHMC Sampler

1

There may be a better way to cache log density and gradient calls in the `drghmc` sampler. At the moment, a stack is used as a cache when we recursively...

gil2rok

During review of PR #39, a request was made to change [the leapfrog implementation in that PR](https://github.com/flatironinstitute/bayes-kit/pull/39/files/b4f75c1973ea6866906f48fdf44c1c9f1339adb7#diff-6c83f04fde917d82f42d704509c142984e42f8e60a6c1feac29289e0c88a22fdR260-R280) (see resolved issues in highlighted section). The end result was subtly different: ```python...

jsoules

Add implementation of ranked R-hat as described in this paper: * Vehtari, A., Gelman, A., Simpson, D., Carpenter, B. and Bürkner, P.C., 2021. [Rank-normalization, folding, and localization: An improved R...

bob-carpenter

I created a local test suite that tries to run `bayes_kit.HMCDiag(..., stepsize=0.01, steps=50)` and `bayes_kit.MALA(epsilon=0.01)` on suite of problems from posteriordb. I am just testing that outputs are not NaN...

wrongu

The goal of this issue is to implement a more robust fix for the issue exposed through PR #28. Many samplers in this package can accept a set of explicit...

jsoules

affine invariant sampler + type cleanup

9

Add affine invariant ensemble sampler. Cleanup typing. 0 mypy errors with all tests passing, with * numpy 1.24.1 * mypy 0.991

bob-carpenter

Testing against known models, use mcse instead of absolute tolerance

1

Consider the scenario of testing a sampler against a model where we know the expectations (e.g. any Normal model). The Metropolis sampler has tests along the lines of ``` model...

roualdes