A new release of pymc-extras is now available!
Release Information
- Version:
v0.15.0 - Repository: pymc-devs/pymc-extras
- Release Page: Release v0.15.0 · pymc-devs/pymc-extras · GitHub
- Note: It may take some time for the release to appear on PyPI and conda-forge.
Release Notes
What’s Changed
Major Changes 
- Deprecate observed=None on Prior.create_likelihood_variable by @anevolbap in #732
- Untangle PytensorRepresentation and assign matrices whole by @jessegrabowski in #738
- Take fit state off the statespace model and make rebuilds work by @jessegrabowski in #741
New Features 
- Add ConvergentFilter by @jessegrabowski in #676
- Stop sharing mutable Kalman filters between graph builds by @jessegrabowski in #735
- New ADVI API by @jessegrabowski in #635
- Fix orthogonalized impulse response functions by @jessegrabowski in #746
- Evaluate stationary VAR likelihoods without a Kalman filter by @jessegrabowski in #766
- Smooth with a Durbin-Koopman disturbance smoother by default by @jessegrabowski in #767
- Add a deterministic trend to BayesianVARMAX and BayesianSARIMAX by @jessegrabowski in #771
- Fold the simulation smoother’s two smoother passes into one by @jessegrabowski in #769
Bugfixes 
- Fix sample_statespace_matrices and give it tests by @jessegrabowski in #734
- Fix AttributeError raised when a placeholder is registered twice by @jessegrabowski in #742
- Fix three cases where ADVI silently ignored the caller by @jessegrabowski in #753
- Fix use_data_time_dim in unconditional sampling by @jessegrabowski in #747
- Fix end and start argument handling in forecast by @jessegrabowski in #748
- Warn when n_steps conflicts with the shock trajectory length by @jessegrabowski in #749
- Fix the normalizing constant in the Kalman filter log-likelihood by @jessegrabowski in #759
- Fix SquareRootFilter treating P0 as a Cholesky factor by @jessegrabowski in #768
Maintenance 
- Support marginalizing DiscreteUniform with batch endpoints and Categorical with symbolic sized P by @ricardoV94 in #724
- Streaming variational inference: out-of-core DataLoader for minibatch ADVI by @YichengYang-Ethan in #698
- Add Simulation Smoother by @jessegrabowski in #680
- Split
statespace.pypost-estimation logic into topical modules by @jessegrabowski in #727 - Clean up dead statespace attributes and fix Representation copy aliasing by @jessegrabowski in #729
- Build post-estimation graphs with the fit’s filter settings by @jessegrabowski in #728
- Deprecate save_kalman_filter_outputs_in_idata and always register data by @jessegrabowski in #691
- Fix DFM state metadata and error covariance construction by @jessegrabowski in #730
- Build the shared-states design matrix explicitly by @jessegrabowski in #737
- Return substituted statespace matrices instead of storing them by @jessegrabowski in #740
- Reworking Minibatching for ADVI by @zaxtax in #713
- Shape logp placeholders per variable and assert output sets exactly by @jessegrabowski in #750
- [FIX] Fix multivariate SquareRootFilter log-likelihood. Targets #744 by @4nmus in #757
New Contributors
- @YichengYang-Ethan made their first contribution in #698
- @4nmus made their first contribution in #757
Full Changelog: Comparing v0.14.0...v0.15.0 · pymc-devs/pymc-extras · GitHub
This post was automatically generated from the GitHub release.