GitLab

Gamma function is parameterized with rate Beta instead of mean Beta

* interpolated plot

* interpolated minor fixes

* minor edits

* histogram fix

* relocate arviz style line position

Co-authored-by: twiecki <twiecki@users.noreply.github.com>

Also see #4536 where the wrappers were brought back for v3.

Closes #4528

* Fix logp of (Discrete) Uniform to not depend on bound
* Add unittest
* Remove redundant `all()` bound conditions in multivariate distributions and improve documentation of dist_math::bound
* Add recommendation for check_bounds.
* Add release note
* Include Release Notes from 3.11.2

It seems like broadcasting information gets lost when applying
`pm.make_shared_replacements`, leading to problems with the metropolis
sampler. Potentially related issues below:
 - https://github.com/pymc-devs/pymc3/issues/1083
 - https://github.com/pymc-devs/pymc3/issues/1304
 - https://github.com/pymc-devs/pymc3/issues/1983

This fix was previously suggested in the following issue:
 - https://github.com/pymc-devs/pymc3/issues/3337

It could be that further adaptations are necessary as indicated in the
issue. Strangely, this does not seem to lead to problems when using
NUTS.

* Add conditional autoregression distribution (CAR)

* Remove float128 dtype which does not exist on Windows and newer OSX versions and furthermore does not to be properly supported even if it does exist.
* Add float128 to release-notes. Add link to memoization PR.
* Remove unused imports.

* Break down tests having multiple checks and xfail decorated

Xfail is going to care only about the first failed assert statement and
ignore all the other ones in the same test.

Sometimes xfail is being used to decorate a test or class function calling
several other sub-functions. This makes it hard to monitor what is happening
for the sub-fuctions that are not failing (given that the failure will take
priority and mark the test function as xfailed). By breaking down each
function into individual sub-functions it's easier to monitor individual
tests behavior

* Remove xfail which seems unnecessary

Also add comment regarding test with unpredictable outcome

* Break down test with multiple sub-checks and xfail where appropriate

* Remove Scipy Xfail from tests not using sp.betabinom

* Add back xfail to potentially failing test when float32 is used

* Temporatily add n_samples=-1 to check tests behavior for all possible parameters

* Remove temporary n_samples and a couple more to do some quick tests

* Remove unnecessary comments

* Replace Xfail with skipif when using Scipy version as condition

Given that betabinom won't exist in previous versions, there should be
no chance for it's behavior to change, given that it won't exist there
is no point in monitoring the outcome of these tests, so it may be worth
skipping them. Skipping them should be faster than running them and
catching the failures

* Add missing n_samples for test_beta_binomial_selfconsistency

* Remove all the n_samples=-1 added for verifying tests robustness

* Tweak normal_logcdf and moyal tests

Temporarily set n_samples=-1 to test all permutations.

* Reintroduce 32bit xfail on Moyal logcdf, revert n_samples and merge passing normal tests

* Small renaming and reordering

* Reintroduce accidentally removed n_samples from binomial test
Co-authored-by: Ricardo <ricardo.vieira1994@gmail.com>

Co-authored-by: Brandon T. Willard <brandonwillard@users.noreply.github.com>

* don't add chain offset in loop already taking care of this

* don't default to chain 0 when computing sampler stats; use provided chain_idx

* adding test script for chain_idx in sample()

* marking test as xfail for now

* take care of chain indices in sample_posterior_predictive

* update test to include sample_posterior_predictive

* use reproducable order of ppc samples wrt multiple chains

* Simplify wald logcdf method

* Return logcdf = 0, when x overflows

* reapplying rename theano -> aesara
Co-authored-by: Eelke Spaak <eelke.spaak@gmail.com>

* Pass stacklevel argument corrently

* Pre-commit errors

* Missing comma

* Modifying cartesian product to allow for more >2D input arrays
* Assert for equality in cartesian test
* Mention #4482 in new features
Co-authored-by: Michael Osthege <michael.osthege@outlook.com>

Closes #4485

* Create bash script for faster build

* Use git-dir in bashscript to change branch

* Better git command for checking out the pymc3-examples branch

* First change the pymc3 examples branch

* Address comments and suggestions on bash script

* Rename bash file according to comments

* Update fast_build_docs.sh

Change `sh` -> `bash` in shell script
Co-authored-by: Marco Gorelli <marcogorelli@protonmail.com>
Co-authored-by: Marco Gorelli <marcogorelli@protonmail.com>

* Small fix Triangular logp and logcdf methods
* Add tests for invalid parameters Uniform, Triangular, DiscreteUniform

* Transform samples from sample_numpyro_nuts

* Add `pymc3.sampling_jax._transform_samples` function which transforms draws

* Modify `pymc3.sampling_jax.sample_numpyro_nuts` function to use this function to return transformed samples

* Add release note

* Update pymc3/sampling_jax.py
Co-authored-by: Junpeng Lao <junpenglao@gmail.com>

* Added a small test

* Split jax tests into their own workflow
Co-authored-by: Michael Osthege <michael.osthege@outlook.com>
Co-authored-by: Junpeng Lao <junpenglao@gmail.com>

* ⬆️ UPGRADE: Autoupdate pre-commit config

* ⬆

️ UPGRADE: Autoupdate pre-commit config
Co-authored-by: twiecki <twiecki@users.noreply.github.com>

* Refactor and unseed TestMatchesScipy
* Show arguments when check_logcdf fails
* Reduce 64bit precision test_skew_normal
* Mark xfail 32-bit for Gamma logcdf
* Reduce 64bit precision test_inverse_gamma_alt_params

* Fix Dirichlet.logp by checking number of categories > 1 only at event dims

* Update test_distributions.py

* Removed the shape validation check to even work for last dimensional shape as 1.

Modified the `test_dirichlet` function to check for the same.

* Added a test to check Dirichlet.logp with different batch shapes.

* Tested exact Dirichlet.logp values againt scipy implementation

Given a mention in RELEASE-NOTES.md

Make deepcopy of start dicts in pm.sample and `pm.find_MAP` to prevent inplace modification of user variables

closes #4456 

* one class docs updated

* Updated with docs with PEP 257 in one class

* Updated with docs with PEP 257 in one class

* updated

* update

* Update

* update

* Update pymc3/step_methods/metropolis.py
Co-authored-by: Thomas Wiecki <thomas.wiecki@gmail.com>

Avoid unclear TypeError when using theano.shared variables as input to distribution parameters (#4445)

* Added default testvalue support for theano.shared
Co-authored-by: Ricardo <ricardo.vieira1994@gmail.com>

* Add tests for edge cases

* Add release-note

* Support imputations with ndarray data (closes #4437)