1.9.1

New Features

• Stable.log_prob() is now implemented and differentiable by BenZickel in #3369, #3370
• Introducing pyro.infer.predictive.WeighedPredictive which reports weights along with predicted samples by @BenZickel in #3345
• Resampler for weighed samples by @BenZickel in #3352
• Add function for calculating quantiles of weighed samples. by @BenZickel in #3340
• Add effective sample size analytics to WeighedPredictive results by @BenZickel in #3351
• Introducing PyroModuleList, because torch.nn.ModueList reinitializies modules when slice-indexing by @MartinBubel in #3339

Type hints

• Type annotate pyro.nn.module in #3337
• Fixes to type hints #3333, #3334

Additional changes

• small update doc HMC by @dario-coscia in #3335
• Type annotate pyro.nn.dense_nn and pyro.nn.auto_reg_nn by @ordabayevy in #3342
• Enable type checking of the tests folder by @ordabayevy in #3336
• Auto regressive nn on gpu - revisited by @cafletezbrant in #3346
• GCC-8 CI error test by @ordabayevy in #3349
• Fix ruff version to 0.3.4 in order to overcome issue with version 0.3.5. by @BenZickel in #3354
• Fix ruff settings & remove flake8 and isort settings by @ordabayevy in #3356
• Unpin ruff version by @BenZickel in #3357
• Switch dependency from jupyter to notebook by @fritzo in #3360
• Add the energy score scoring rule for comparing multivariate distributions by @BenZickel in #3362
• Add jupyter lab option to docker Makefile by @eteq in #3363
• Replace torch._utils._accumulate with numpy.cumsum. by @BenZickel in #3364
• Fix PyroModule rendering error in local-parameter mode by @eb8680 in #3366

New Contributors

• @dario-coscia made their first contribution in #3335
• @MartinBubel made their first contribution in #3339
• @BenZickel made their first contribution in #3340
• @cafletezbrant made their first contribution in #3346
• @eteq made their first contribution in #3363

Full Changelog: 1.9.0...1.9.1

1.9.0

Breaking Changes

• Drop support for PyTorch 1
• Drop support for Python 3.7

New features

• Type hints are now available in some parts of Pyro, thanks to @ordabayevy. Please let us know about any issues. We welcome help adding type hints to remaining parts of Pyro. #2550
• New tutorials using Zuko normalizing flows by @francois-rozet
• A new simple RandomWalkKernel by @martinjankowiak
• A new SimplexToOrderedTransform by @peblair
• A new simple PyTorch training example by @eb8680

Detailed Changes

• Fix GammaPoisson distribution from ignoring validate_args param by @austinv11 in #3253
• Updated docs and fixed spelling by @martinrohbeck in #3254
• fix custom_guide in intro_long.ipynb by @martinjankowiak in #3258
• use generator arg in dataloader in cevae by @martinjankowiak in #3264
• SBVM sampler fix by @deoxyribose in #3261
• Render deterministic functions by @r3v1 in #3266
• Replace lap.lapjv() with scipy.optimize.linear_sum_assignment() by @fritzo in #3267
• Fix: Wrong codomain in LeakyReLUTransform by @MeenuyD in #3270
• Drop the support for PyTorch<2.0 by @ordabayevy in #3272
• Add param_store.py type hints by @ordabayevy in #3271
• Address the issues in #3274 by @cyianor in #3277
• Drop Python 3.7 support by @ordabayevy in #3278
• Update link to Nolan reference of Stable distribution by @fehiepsi in #3281
• Fix pyro.poutine.handlers by @ordabayevy in #3283
• Add handlers explicitly by @ordabayevy in #3285
• Use functools.wraps to preserve handler signature by @ordabayevy in #3287
• Type annotate pyro.poutine.runtime by @ordabayevy in #3288
• Type annotate pyro.poutine.messenger by @ordabayevy in #3290
• Type annotate pyro.primitives & poutine.block_messenger by @ordabayevy in #3292
• Add os to readthedocs.yml file by @fehiepsi in #3293
• Remove python.version entry in rtd.yml file by @fehiepsi in #3294
• Type annotate broadcast, condition, do, enum messengers by @ordabayevy in #3295
• Typo in the documentation example by @pavleb in #3298
• Type annotate Trace, TraceMessenger, & pyro.poutine.guide by @ordabayevy in #3299
• Small typo in tutorial by @osorensen in #3300
• Update effect_handlers.ipynb by @djinnome in #3296
• Fix gate & gate_logits in ZeroInflatedDistribution by @ordabayevy in #3303
• Type annotate poutines by @ordabayevy in #3306
• Numpy Import Convention by @RatishT in #3307
• Type annotate messengers by @ordabayevy in #3308
• Add simple RandomWalkKernel by @martinjankowiak in #3311
• Type annotate messengers by @ordabayevy in #3309
• Add tutorials using normalizing flows by @francois-rozet in #3302
• Mypy warn_unreachable=True by @ordabayevy in #3312
• Fix mypy and line-too-long errors by @fritzo in #3313
• Get CI working with new black, torch, pytest by @fritzo in #3318
• Annotate handlers & add py.typed by @ordabayevy in #3321
• Port implementation of SimplexToOrderedTransform from numpyro by @peblair in #3320
• Prevent observation masks improperly expanding samples in plates by @austinv11 in #3317
• Document InferDict and Message typed dicts by @fritzo in #3323
• Fix bug in CVAE example by @FDUguchunhui in #3325
• Add a simple PyTorch training example by @eb8680 in #3327
• Add .github/workflows/publish.yml to publish to pypi.org by @fritzo in #3326
• Fix interaction between PyroParam and torch.func.grad by @eb8680 in #3328
• Fix action link for publishing to PyPI by @fehiepsi in #3329
• Bump version to 1.9.0 by @fritzo in #3330

New Contributors

• @austinv11 made their first contribution in #3253
• @deoxyribose made their first contribution in #3261
• @r3v1 made their first contribution in #3266
• @MeenuyD made their first contribution in #3270
• @cyianor made their first contribution in #3277
• @pavleb made their first contribution in #3298
• @osorensen made their first contribution in #3300
• @djinnome made their first contribution in #3296
• @RatishT made their first contribution in #3307
• @peblair made their first contribution in #3320
• @FDUguchunhui made their first contribution in #3325

Full Changelog: 1.8.6...1.9.0

1.8.6

Summary

• Reenable support for PyTorch 1.11 (after Pyro's 1.8.5 release had narrowly required torch>=2.0)
• Improve docstrings of Pyro's distribution wrappers around torch distributions #3246

What's Changed

• Refactor ProvenanceTensor to use pytree by @ordabayevy in #3223
• Create FUNDING.yml by @fritzo in #3224
• Remove DataParallel container in SS-VAE model by @martinrohbeck in #3227
• Fix jit ELBO sign error by @fritzo in #3229
• Replace flake8+isort with ruff by @fritzo in #3231
• fix _loc_scale method in AutoMultivariateNormal by @martinjankowiak in #3233
• Revert to support torch>=1.11.0 by @francois-rozet in #3242
• Concatenate torch docstrings at end of distribution wrappers by @fritzo in #3246
• Disable pdf builds on readthedocs by @fritzo in #3250
• Disable pdf and epub build on readthedocs by @fehiepsi in #3251
• Fix #3245 bug in poutine.trace(graph_type="dense") by @fritzo in #3247
• Bump version to to 1.8.6 by @fritzo in #3252

New Contributors

• @francois-rozet made their first contribution in #3242
• @rtviii collaborated with @fritzo in #3246

Full Changelog: 1.8.5...1.8.6

1.8.5

This release includes a number of fixes to support PyTorch 2.

What's Changed

• Update base learning rate scheduler by @eb8680 in #3167
• Update GP tutorials to unblock #3174 by @fehiepsi in #3175
• add substitute handler by @thisiscam in #3125
• Update torch_patch.py by @S163669 in #3174
• fix: typo in log-absolute-derivative of exponential transform (normalizing flow tutorial) closes #3176 by @felixschltr in #3177
• rerun newest black on entire repo by @martinjankowiak in #3178
• Preserve path to the file when rendering model by @LysSanzMoreta in #3186
• Add conditional inverse and compose TransformModules by @eb8680 in #3185
• PyTorch Lightning example by @ordabayevy in #3189
• Fix & update GMM tutorial by @ordabayevy in #3193
• Fixing typos in infer.inspect by @ordabayevy in #3198
• Bug fix about device of subsample in subsample_messenger.py by @hjnnjh in #3195
• Dependency tracking tutorial by @ordabayevy in #3197
• Make patch_dependency safer under dependency drift by @fritzo in #3204
• Deprecate CorrLCholeskyTransform in favor of upstream CorrCholeskyTransform by @ordabayevy in #3199
• VAE: Improved ELBO plots by @martinrohbeck in #3205
• Remove vestigial lines from pyro/init.py all by @fritzo in #3210
• fix learning rate schedulers in PyTorch 2.0 (closes #3202) by @ilia-kats in #3207
• Clear .unconstrained weakrefs before pickling; rebuild them more often by @fritzo in #3212
• Add pl.Trainer args to argparse by @ordabayevy in #3217
• Run CI with PyTorch 2.0 by @eb8680 in #3192
• make ProvenanceTensor behave more like a Tensor (closes #3218) by @ilia-kats in #3220
• Bump version to 1.8.5 by @fritzo in #3222

New Contributors

• @thisiscam made their first contribution in #3125
• @S163669 made their first contribution in #3174
• @felixschltr made their first contribution in #3177
• @LysSanzMoreta made their first contribution in #3186
• @hjnnjh made their first contribution in #3195
• @martinrohbeck made their first contribution in #3205
• @ilia-kats made their first contribution in #3207

Full Changelog: 1.8.4...1.8.5

1.8.4

Fixed

• Allow torch>=2 by @fritzo in #3164
• fixed NaN with in log_prob corr < -1e-8 for SineBivariateVonMises by @OlaRonning in #3165
• Allow registering of custom exception handlers for potential_fn computations by @Balandat in #3168

Added

• Allow rendering of semisupervised models by @fritzo in #3159
• Add GroupedNormalNormal distribution by @martinjankowiak in #3163
• Add make tutorial to docs CI workflow by @eb8680 in #3012

New Contributors

• @Balandat made their first contribution in #3168

Full Changelog: 1.8.3...1.8.4

1.8.3

What's Changed

• rename custom_objectives_training.ipynb -> custom_objectives.ipynb by @martinjankowiak in #3141
• Expose Gaussian algorithms by @fritzo in #3145
• fix_reverse_out_bound_quadratic_spline by @LiaoShiqi97 in #3140
• Improve sequential_gaussian_filter_sample() by @fritzo in #3146
• Add jitter to Cholesky factorization in Gaussian ops by @fritzo in #3151
• Clean up handling of global settings by @fritzo in #3152
• Add an option to stop PyroModules from sharing parameters by @eb8680 in #3149
• Ensure compatibility with torch>=1.13 torchvision>=0.14 by @fritzo in #3155
• Adds min_stepsize and max_stepsize to HMC, avoiding infinite loop by @fritzo in #3157

New Contributors

• @LiaoShiqi97 made their first contribution in #3140

Full Changelog: 1.8.2...1.8.3

1.8.2

What's Changed

• Fix nbshpinx rendering of tutorials by @fritzo in #3055
• Modified docker file to use wget instead of curl by @Jayanth-kumar5566 in #3058
• Updating Docker file to use Ubuntu 18.04 by @Jayanth-kumar5566 in #3059
• Updated CUDA and base OS version by @Jayanth-kumar5566 in #3060
• Fix contrib.funsor.Trace_EnumELBO model enumeration by @ordabayevy in #3063
• Fix docstrings of distributions rendered on sphinx by @fehiepsi in #3064
• Add test applying AutoNormal to two different models by @fritzo in #3065
• Fix link in AIR tutorial by @dilaragokay in #3066
• Numerically stabilize ProjectedNormal.log_prob() via erfc by @fritzo in #3071
• fix documentation class names to VariationalSparseGP by @ivetasarfyova in #3076
• fix Nystrom approximation equation indices by @ivetasarfyova in #3077
• remove extraneous svi step call in documentation by @ivetasarfyova in #3078
• Add type assertion for model_args and model_kwargs of render_model by @dilaragokay in #3083
• Remove batch_expand helper function in air example by @ordabayevy in #3086
• Fix a typo in air tutorial by @ordabayevy in #3087
• Fix Nyström Typo by @adiehl96 in #3084
• Use provenance tracking to compute downstream costs in TraceGraph_ELBO by @ordabayevy in #3081
• Use : for param domain in render_model by @fehiepsi in #3097
• store batch_shape as torch.Size object in MultivariateStudentT distribution by @flo-schu in #3099
• Fix bound partial for python 3.10 by @fehiepsi in #3101
• Fix _logmatmulexp comment by @fritzo in #3105
• Fixing typos in enumeration tutorial by @ordabayevy in #3109
• Expose docs for pyro.contrib.funsor.infer.infer_discrete by @fritzo in #3111
• Update signature of sample in tutorial intro_long by @fraterenz in #3112
• Allow autocorrelation() to run without MKL by @fritzo in #3113
• Fix multinomial_goodness_of_fit printing by @fritzo in #3115
• Numerically stabilize autocorrelation() by @fritzo in #3114
• Implement NanMaskedNormal, NanMaskedMultivariateNormal by @fritzo in #3116
• Fix links in workflow tutorial by @fritzo in #3119
• Implement Resamper for interactive prior tuning by @fritzo in #3118
• fix entropy term in distributions.py (wrong isinstance check) by @martinjankowiak in #3120
• Custom loss documentation by @e-pet in #3122
• Fix model_7 in examples/contrib/funsor/hmm.py by @ordabayevy in #3126
• Fix memory leak in TraceEnumELBO by @fehiepsi in #3131
• Bump version to 1.8.2 by @fritzo in #3135

New Contributors

• @Jayanth-kumar5566 made their first contribution in #3058
• @dilaragokay made their first contribution in #3066
• @ivetasarfyova made their first contribution in #3076
• @adiehl96 made their first contribution in #3084
• @flo-schu made their first contribution in #3099
• @fraterenz made their first contribution in #3112
• @e-pet made their first contribution in #3122

Full Changelog: 1.8.1...1.8.2

1.8.1

Breaking changes

• Update to PyTorch 1.11.0 in #3045
• Drop support for Python 3.6 in #3047

New features

• New tutorial on Bayesian workflow, with a SARS-CoV-2 running example in #2977
• Render parameters in render_model() @karm-patel in #3039
• Distributions:
  • Add LogNormalNegativeBinomial distribution by @martinjankowiak in #3010
  • Implement DiscreteHMM.sample() by @fritzo in #3050
  • Parallelized DiscreteHMM.sample() by @ordabayevy in #3053
  • Add missing docs for ProjectedNormal parameter in #3009
  • Support 4D ProjectedNormal distribution in #3011

Misc changes

• fix docstring in repeated_matmul by @martinjankowiak in #2999
• use more conservative learning rate in bart.py by @martinjankowiak in #3002
• Clarified comments in minipyro.py to fix #3003 by @luiarthur in #3004
• Update black formatting by @fritzo in #3020
• Closes #3016 by @nipunbatra in #3017
• Fix timeseries tutorial link by @fritzo in #3021
• Some documentation improvements to MLE_MAP and SVI_P by @nipunbatra in #3022
• Improvements to GP tutorial by @nipunbatra in #3027
• change sample to param statements in guides by @dhudsmith in #3042
• Added type assertion for better code clarity by @GautamV234 in #3036
• add GitHub URL for PyPi by @andriyor in #3038

New Contributors

• @luiarthur made their first contribution in #3004
• @nipunbatra made their first contribution in #3017
• @dhudsmith made their first contribution in #3042
• @GautamV234 made their first contribution in #3036
• @karm-patel made their first contribution in #3039
• @andriyor made their first contribution in #3038

Full Changelog: 1.8.0...1.8.1

1.8.0 release

New features and improvements

• New Tutorials
  • Introduction to Pyro, a completely new intro tutorial #2991
  • Logistic growth models of SARS-CoV-2, #2982
  • SCANVI for single cell transcriptomics, #2975
  • Model rendering #2962
• Added pyro.render_model() to visualize model structure #2962 (based on @ordabayevy's ProvenanceTensor and numpyr.render_model())
• Autoguides
  • Improve parametrization of AutoMultivariateNormal #2972, #2963
  • Added a new class of effect-based guides including AutoNormalMessenger, AutohierarchicalNormalMessenger, and AutoRegressiveMessenger #2953, #2955 (thanks to @vitkl)
  • Added an AutoGaussian, similar to AutoMultivariateNormal, but with sparse precision matrix factorization based on dependency structure in the model.
• Added sampling functions to the MuE missing data discrete HMM #2898
• Added a PositivePowerTransform for positive-valued variables #2904

1.7.0 release

New features

• Update to PyTorch 1.9 #2887
• A StreamingMCMC class for high-dimensional Bayesian inference using NUTS or HMC, thanks to @mtsokol #2857 .
  StreamingMCMC is a drop-in replacement for MCMC that avoids storing samples during inference by streamingly computing statistics such as mean, variance, and r_hat. You can define your own statistics using the pyro.ops.streaming module by either composing existing statistics or defining your own subclass of StreamingStats #2856 .
• Make poutine.reparam compatible with initialization logic in autoguides and MCMC #2876 . Previously you needed to manually transform the value in init_to_value() when using a reparametrizer. In Pyro 1.7 you can specify a single init_to_value() output that should work regardless of whether your model is transformed by a reparametrizer. Note this involves a major refactoring of the Reparam interface, namely replacing .call() with .apply(). If you have defined custom reparametrizers using .__call__() you should refactor them before the next Pyro release.
• Add an AutoStructured guide with flexible distributions, sparse flexible dependencies among latent variables, and support for reparametrization #2812 . This autoguide is somewhere between AutoNormal and AutoIAFNormal+NeuTraReparam. Like AutoNormal this guide is interpretable and structured. Like NeuTraReparam this guide is flexible and can be used to improve geometry for subsequent inference via HMC or NUTS.
• New directional distributions thanks to @OlaRonning
  • SinSkewedToroidal #2826
  • SinBivariateVonMises #2821
• New distributions over genetic and amino acid sequences, thanks to @EWeinstein #2728
  (see https://www.biorxiv.org/content/10.1101/2020.07.31.231381v2)
• New distributions with exponential tails
  • Logistic #2895
  • SkewLogistic #2874
  • AsymmetricLaplace #2850
  • SoftLaplace #2791
  • SoftAsymmetricLaplace #2872
• Add AutoGuideList.quantiles() thanks to @vitkl #2896
• Allow saving a subset of variables in MCMC via the save_params option, which can save memory #2793
• Add a sample option to pyro.contrib.funsor.infer_discrete #2789

New tutorials

• A predictive and deterministic tutorial thanks to @mtsokol #2852
• ProfileHMM with MuE by @EWeinstein #2728
• Factor MuE and Probabilistic PCA by @EWeinstein #2728
• Add @ordabayevy's trick to the enumeration tutorial #2892
• Enhance the LDA tutorial #2790

Bug fixes

• Fix poutine.do to avoid duplicate entries in cond_indep_stack #2846
• Fix infer.csis to ignore unused gradients, thanks to @fshipy #2828
• Catch NAN values in TraceENUM_ELBO grads #2782

Breaking changes

• Simplify param names in callable optim configs #2814
• Switch to softplus transform when using autoguide scales, thanks to experiments performed by @vitkl #2823

Ceanup

• Start using mypy for type checking, thanks to @kamathhrishi #2853 #2858
• Start using black code formatter #2891