0.12.1

This release includes a fix for jax 0.4.11 (#1595).

0.12.0

New Features

• New distribution: Gompertz distribution (#1551)
• New initialization strategy: init_to_mean (#1550)
• New examples and tutorials:
  • Illustrate the usage of JAX PositionalSharding for distributing the computations of log_prob/grad over multiple devices in MCMC (#1514)
  • A port of Gaussian Mixture Model tutorial from Pyro (#1562)
  • A toy mixture model with discrete enumeration (#1568)
• New inference utilities get_transforms and unconstrain_fn to transform between unconstrained and constrained space (#1564)
• Support jaxns>=2.0.1 (#1546)

Enhancements and Bug Fixes

• Make transforms jittable (#1575)
• Fixed typo in surrogate posterior of beta (#1591)
• Do not scale mnist label (#1589)
• Do not mutate shapes of ExpandedDistribution for map-free ops (#1574)
• Add support for JAX custom PRNG (#1587)
• Include deterministic variables in AutoDelta's sample_posterior (#1584)
• Fix forward shape of SimplexToOrderTransform (#1583)
• Fix inf's in TruncatedNormal log_prob & sample (#1581)
• Allow users to specify total_count_max in Multinomial (#1557)
• Allow pickled mcmc object to run post warmup phase (#1558)
• Add init_params argument to svi.init() and svi.run() (#1561)
• Support pickling MCMC objects with enumeration (#1577)
• Raise error when reparameterize lognormal (#1548)
• Avoid initializing model params when already specified in guide (#1553)
• Respect init params if provided to mcmc.run (#1547)
• Fix provenance for jax 0.4.4 (#1543)
• Use analytic kl divergence in TraceEnum_ELBO (#1533)
• Properly handle contraction of guide plates in TraceEnum_ELBO (#1537)
• Raise an error if there is no common scale when model enumerated (#1536)
• Optimize reduction of enumerated guide sites (#1531)
• Guess max_plate_nesting in TraceEnum_ELBO (#1528)

0.11.0

Breaking changes

Drop Python 3.7 support and require the minimal jax version 0.4.

New Features

• New distributions:
  • EulerMaruyama for modelling stochastic differential equations (SDE) (thanks to @yayami3)
  • GaussianCopula and GaussianCopulaBeta (i.e. MultivariateBeta) to incorporate correlations into univariate random variables (thanks to @hessammehr)
  • MatrixNormal generalizes multivariate normal to matrix-valued random variables (thanks to @kaijennissen)
  • LogUniform distribution (thanks to @yayami3 and @andrewfowlie)
• New objective TraceEnum_ELBO to support enumeration in SVI (thanks to @ordabayevy)
• New tutorial:
  • Text-Based Ideal Points using NumPyro (thanks to @elchorro)
• #1508 Allow rendering deterministic sites (thanks to @YanniPapandreou)

Enhancements and Bug Fixes

• #1507 Made constraints robust to pickling (thanks to @pierreglaser)
• #1515 Fixed log_prob for negative correlation in SineBivariateVonMises distribution (thanks to @OlaRonning)
• #1509 Fixed a bug at flatten/unflatten distributions which mixed the order of arguments of the distributions (thanks to @hessammehr)
• #1494 Fixed Mixture distribution with unnormalized logits giving incorrect log_prob
• #1486 Returned a positive ordered vector when applies ExpTransform for an ordered vector
• #1491 Fixed Mixture intermediate values
• #1480 Fixed some computations in Bayesian Hierarchical Stacking tutorial (thanks to @cpieringer)
• #1478 Added icdf methods for Beta, Gamma, StudentT
• #1477 Allowed multiple arguments to initialize flax/haiku modules (thanks to @fehiepsi)
• #1475 Used TFP's betainc (which supports taking gradient w.r.t. parameters) in StudentT.cdf (thanks to @colehaus)

0.10.1

New Features

• #1458 New Mixture distribution to allow component distributions come from different families (thanks to @dfm)

Enhancements and Bug Fixes

• #1469 Fix memory leak when using scan
• #1465 Fix validation logic for transformed distribution (thanks to @xidulu)
• #1462 Avoid infinite loop in vmapped Binomial (thanks to @tbenthompson)
• #1449 Move contrib.render functionality to infer.inspect (thanks to @mattja)

0.10.0

New Features

• #1394 New distribution Conditional Autoregressive CAR (thanks to @theorashid)
• #1434 New flexible auto guides for models with both global and local latent variables: AutoSemiDAIS and AutoSurrogateLikelihoodDAIS
• #1429 New example: Conditional Variational Autoencoder in Flax (thanks to @dirmeier)

Enhancements and Bug Fixes

• #1401 Fix obs argument is not respected when sample primitive is not executed under any handler (thanks to @gcskoenig)
• #1412 TraceGraph_ELBO implementation using provenance tracking
• #1418 Fix SA sampler cannot be run in parallel chains
• #1419 Fix categorical sampler occasionally generate out-of-support samples
• #1436 Allow to use potential_fn in BarkerMH
• #1437 Fix for AutoMultivariateNormal.get_posterior method return incorrect distribution (thanks to @xidulu)
• #1444 Promote shapes for observed variables inside scan's transition function
• #1443 Consider the time dimension of markov models with history=0 as plate
• #1441 More stable check for corr_cholesky constraint
• #1400 Fix SineBivariateVonMises sampler

0.9.2

New Features

• #1381 New render_params argument for render_model
• #1366 Allow a fixed number of steps in HMC.
• New tutorials and examples:
  • Modelling mortality over space and time
  • New section on adding intermediate levels to Bayesian Hierarchical Linear Regression tutorial

Enhancements and Bug Fixes

• #1386 Support JAX 0.3.5
• #1388 Update jaxns to 1.0.0
• #1372 Fix batch shapes of SineBivariateVonMises distribution
• #1375 Add __repr__ method for constraints
• #1358 Force validate_args to be keyword argument
• #1350 Enhance the AR2 example

Thanks, @cstoafer, @hesenp, @tcbegley, @themrzmaster, @karm-patel, @theorashid

0.9.1

New Features

• New distributions: AsymmetricLaplace and AsymmetricLaplaceQuantile for quantile regression.
• New tutorials and examples:
  • Bayesian neural network with SteinVI
  • Deep markov model inferred using SteinVI
  • Zero-Inflated Poisson regression model

Enhancements and Bug Fixes

• #1320 Fix provenance logic for numpyro.render_model
• #1330 Fix vectorize sampling for sites with size=0
• #1331 Allow nested params in SteinVI
• #1333 Reduce memory consumption for SteinVI
• #1325 Fix compat MCMC signature

Thanks, @Vinnie-Palazeti, @wataruhashimoto52, @hessammehr, @OlaRonning, @d-diaz!

0.9.0

New Features

• New VI inference: SteinVI. Checkout a couple examples in PRs #1297 #1298 for the usage.
• New distributions: MultivariateStudentT, DiscreteUniform, Kumaraswamy, RelaxedBernoulli .
• New tutorials and examples:
  • Tutorial for Truncated distributions: a complete guide for how to construct a NumPyro distribution.
  • Bayesian Hierarchical Stacking case study to average models based on weights from a hierarchical structure.
  • Sine-skewed sine (bivariate von Mises) mixture to model the dihedral angles that occur in the backbone of a protein.
  • AR2 processes to show how to avoid the (slow) Python for-loop.
  • Holt-winter Exponential Smoothing example for time series forecasting.
  • Hilbert space approximation for Gaussian processes example is significantly revised.

Enhancements and Bug Fixes

• #1305 Fixes HMCECS bug for likelihoods with multiple plates
• #1304 Improves warning mechanism when plates are missing.
• #1301 Fixes sparse Poisson density sometimes returns int output.
• #1289 Make HMC Gibbs algorithms work with improper distributions
• #1284 Adds various KL divergences for Gamma/Beta families
• #1281 Raises error if there are duplicated deterministic sites
• #1271 Better warning mechanism with stacklevel
• #1270 Incorporate kl divergences of Tensorflow Probability distributions
• #1259 #1266 Allow TruncatedNormal/Cauchy to take both low and high
• #1254 numpyro.contrib.indexing is moved to numpyro.ops.indexing
• #1252 Use multipledispatch for kl_registry
• #1250 Added cdf methods for gamma, inverse gamma, log normal densities
• #1248 Add ProvenanceArray to infer relational structure in a model
• #1244 Raise warning for the automatic enumeration behavior
• #1237 Enhance warnings for invalid parameters of BetaProportion distribution
• #1227 Allow prior to be callable in random_flax_module and random_haiku_module
• #1226 Allow init_to_sample work with scalar values
• #1225 Add color for divergences in Neal's example
• #1196 Allow custom precision function in laplace approximation autoguide
• #1194 Option to specify init state for SVI run
• #1185 #1189 Avoid -inf/nan samples in truncated distributions
• #1182 Extend scope handler for plate stack frames
• #1179 Support enumerate support for zero inflated distributions
• #1169 Allow pickle autoguides

This release is composed of great contributions and feedback from the Pyro community: @amalvaidya @MarcoGorelli @omarfsosa @maw501 @bjeffrey92 @hessammehr @OlaRonning @dykim29 @Carlosbogo @wataruhashimoto52 @Vedranh13 @ahmadsalim @austereantelope and many others. Thank you!

0.8.0

Breaking changes

Switch to softplus transforms for autoguide scales (thanks to experiments performed by @vitkl).

New Features

• New autoguide: AutoDAIS leverages HMC and annealed importance sampling within a variational inference framework
• New distributions: MixtureSameFamily, and directional distributions SineBivariateVonMises, SineSkewed
• New constraints: l1_ball for vectors with L1 norm less than 1
• New transforms: L1BallTransform, SimplexToOrderedTransform, ScaledUnitLowerCholeskyTransform
• #1116 New format_shapes utility to interpret the shapes of random variables/plates in a model.
• #1109 Allow direct use of TFP distributions in numpyro.sample
• New tutorials and examples:
  • Principled prior with Dirichlet distribution for Ordinal Regression case study
  • Horseshoe regression
  • Bad posterior geometry and how to deal with it

Enhancements and Bug Fixes

• #1108 Avoid numerical problems when using BernoulliProbs
• #1118 Recommend AutoNormal guide when hessian in AutoLaplace is singular
• #1126 Smarter warning about discrete inference in SVI models
• #1136 Support to use SA sampler with arviz
• #1139 Document Poisson is_sparse argument
• #1140 Make Sigmoid and StickBreakingTransform more stable
• #1149 Raise value error if num_steps bad in svi.run
• #1162 Use black[jupyter] in notebooks

This release is composed of great contributions and feedback from the Pyro community: @MarcoGorelli @OlaRonning @d-diaz @quattro @svilupp @peterroelants @prashjet @freddyaboulton @tcbegley @julianstastny @alexlyttle and many others. Thank you!

0.7.2

This is a patch release with the following new feature and fixes:

• New example Hilbert space approximation of Gaussian processes #1097 thanks to @omarfsosa
• Fix for rendering models with only discrete variables #1099 thanks to @bdatko
• Fix progress-bar issues when running multi-chain MCMC #1101