Skip to content

Latest commit

 

History

History
291 lines (212 loc) · 11.1 KB

CHANGELOG.md

File metadata and controls

291 lines (212 loc) · 11.1 KB
Raw

v0.18.0

Breaking changes

  • Posteriors saved under sbi v0.17.2 or older can not be loaded under sbi v0.18.0 or newer.
  • sample_with can no longer be passed to .sample(). Instead, the user has to rerun .build_posterior(sample_with=...). (#573)
  • the posterior no longer has the the method .sample_conditional(). Using this feature now requires using the sampler interface (see tutorial here) (#573)
  • retrain_from_scratch_each_round is now called retrain_from_scratch (#598, thanks to @jnsbck)
  • API changes that had been introduced in sbi v0.14.0 and v0.15.0 are not enforced. Using the interface prior to those changes leads to an error (#645)
  • prior passed to SNPE / SNLE / SNRE must be a PyTorch distribution (#655), see FAQ-7 for how to pass use custom prior.

Major changes and bug fixes

  • new sampler interface (#573)
  • posterior quality assurance with simulation-based calibration (SBC) (#501)
  • added Sequential Neural Variational Inference (SNVI) (Glöckler et al. 2022) (#609, thanks to @manuelgloeckler)
  • bugfix for SNPE-C with mixture density networks (#573)
  • bugfix for sampling-importance resampling (SIR) as init_strategy for MCMC (#646)
  • new density estimator for neural likelihood estimation with mixed data types (MNLE, #638)
  • MCMC can now be parallelized across CPUs (#648)
  • improved device check to remove several GPU issues (#610, thanks to @LouisRouillard)

Enhancements

  • pairplot takes ax and fig (#557)
  • bugfix for rejection sampling (#561)
  • remove warninig when using multiple transforms with NSF in single dimension (#537)
  • Sampling-importance-resampling (SIR) is now the default init_strategy for MCMC (#605)
  • change mp_context to allow for multi-chain pyro samplers (#608, thanks to @sethaxen)
  • tutorial on posterior predictive checks (#592, thanks to @LouisRouillard)
  • add FAQ entry for using a custom prior (#595, thanks to @jnsbck)
  • add methods to plot tensorboard data (#593, thanks to @lappalainenj)
  • add option to pass the support for custom priors (#602)
  • plotting method for 1D marginals (#600, thanks to @guymoss)
  • fix GPU issues for conditional_pairplot and ActiveSubspace (#613)
  • MCMC can be performed in unconstrained space also when using a MultipleIndependent distribution as prior (#619)
  • added z-scoring option for structured data (#597, thanks to @rdgao)
  • refactor c2st; change its default classifier to random forest (#503, thanks to @psteinb)
  • MCMC init_strategy is now called proposal instead of prior (#602)
  • inference objects can be serialized with pickle (#617)
  • preconfigured fully connected embedding net (#644, thanks to @JuliaLinhart #624)
  • posterior ensembles (#612, thanks to @jnsbck)
  • remove gradients before returning the posterior (#631, thanks to @tomMoral)
  • reduce batchsize of rejection sampling if few samples are left (#631, thanks to @tomMoral)
  • tutorial for how to use SBC (#629, thanks to @psteinb)
  • tutorial for how to use SBI with trial-based data and mixed data types (#638)
  • allow to use a RestrictedPrior as prior for SNPE (#642)
  • optional pre-configured embedding nets (#568, #644, thanks to @JuliaLinhart)

v0.17.2

Minor changes

  • bug fix for transforms in KDE (#552)

v0.17.1

Minor changes

  • improve kwarg handling for rejection abc and smcabc
  • typo and link fixes (#549, thanks to @pitmonticone)
  • tutorial notebook on crafting summary statistics with sbi (#511, thanks to @ybernaerts)
  • small fixes and improved documenentation for device handling (#544, thanks to @milagorecki)

v0.17.0

Major changes

  • New API for specifying sampling methods (#487). Old syntax:
posterior = inference.build_posterior(sample_with_mcmc=True)

New syntax:

posterior = inference.build_posterior(sample_with="mcmc")  # or "rejection"
  • Rejection sampling for likelihood(-ratio)-based posteriors (#487)
  • MCMC in unconstrained and z-scored space (#510)
  • Prior is now allowed to lie on GPU. The prior has to be on the same device as the one passed for training (#519).
  • Rejection-ABC and SMC-ABC now return the accepted particles / parameters by default, or a KDE fit on those particles (kde=True) (#525).
  • Fast analytical sampling, evaluation and conditioning for DirectPosterior trained with MDNs (thanks @jnsbck #458).

Minor changes

  • scatter allowed for diagonal entries in pairplot (#510)
  • Changes to default hyperparameters for SNPE_A (thanks @famura, #496, #497)
  • bugfix for within_prior checks (#506)

v0.16.0

Major changes

  • Implementation of SNPE-A (thanks @famura and @theogruner, #474, #478, #480, #482)
  • Option to do inference over iid observations with SNLE and SNRE (#484, #488)

Minor changes

  • Fixed unused argument num_bins when using nsf as density estimator (#465)
  • Fixes to adapt to the new support handling in torch v1.8.0 (#469)
  • More scalars for monitoring training progress (thanks @psteinb #471)
  • Fixed bug in minimal.py (thanks @psteinb, #485)
  • Depend on pyknos v0.14.2

v0.15.1

  • add option to pass torch.data.DataLoader kwargs to all inference methods (thanks @narendramukherjee, #445)
  • fix bug due to release of torch v1.8.0 (#451)
  • expose leakage_correction parameters for log_prob correction in unnormalized posteriors (thanks @famura, #454)

v0.15.0

Major changes

  • Active subspaces for sensitivity analysis (#394, tutorial)
  • Method to compute the maximum-a-posteriori estimate from the posterior (#412)

API changes

  • pairplot(), conditional_pairplot(), and conditional_corrcoeff() should now be imported from sbi.analysis instead of sbi.utils (#394).
  • Changed fig_size to figsize in pairplot (#394).
  • moved user_input_checks to sbi.utils (#430).

Minor changes

  • Depend on new joblib=1.0.0 and fix progress bar updates for multiprocessing (#421).
  • Fix for embedding nets with SNRE (thanks @adittmann, #425).
  • Is it now optional to pass a prior distribution when using SNPE (#426).
  • Support loading of posteriors saved after sbi v0.15.0 (#427, thanks @psteinb).
  • Neural network training can be resumed (#431).
  • Allow using NSF to estimate 1D distributions (#438).
  • Fix type checks in input checks (thanks @psteinb, #439).
  • Bugfix for GPU training with SNRE_A (thanks @glouppe, #442).

v0.14.3

  • Fixup for conditional correlation matrix (thanks @JBeckUniTb, #404)
  • z-score data using only the training data (#411)

v0.14.2

  • Small fix for SMC-ABC with semi-automatic summary statistics (#402)

v0.14.1

  • Support for training and sampling on GPU including fixes from nflows (#331)
  • Bug fix for SNPE with neural spline flow and MCMC (#398)
  • Small fix for SMC-ABC particles covariance
  • Small fix for rejection-classifier (#396)

v0.14.0

  • New flexible interface API (#378). This is going to be a breaking change for users of the flexible interface and you will have to change your code. Old syntax:
from sbi.inference import SNPE, prepare_for_sbi

simulator, prior = prepare_for_sbi(simulator, prior)
inference = SNPE(simulator, prior)

# Simulate, train, and build posterior.
posterior = inference(num_simulation=1000)

New syntax:

from sbi.inference import SNPE, prepare_for_sbi, simulate_for_sbi

simulator, prior = prepare_for_sbi(simulator, prior)
inference = SNPE(prior)

theta, x = simulate_for_sbi(simulator, proposal=prior, num_simulations=1000)
density_estimator = inference.append_simulations(theta, x).train()
posterior = inference.build_posterior(density_estimator)  # MCMC kwargs go here.

More information can be found here here.

  • Fixed typo in docs for infer (thanks @glouppe, #370)
  • New RestrictionEstimator to learn regions of bad simulation outputs (#390)
  • Improvements for and new ABC methods (#395)
    • Linear regression adjustment as in Beaumont et al. 2002 for both MCABC and SMCABC
    • Semi-automatic summary statistics as in Fearnhead & Prangle 2012 for both MCABC and SMCABC
    • Small fixes to perturbation kernel covariance estimation in SMCABC.

v0.13.2

  • Fix bug in SNRE (#363)
  • Fix warnings for multi-D x (#361)
  • Small improvements to MCMC, verbosity and continuing of chains (#347, #348)

v0.13.1

  • Make logging of vectorized numpy slice sampler slightly less verbose and address NumPy future warning (#347)
  • Allow continuation of MCMC chains (#348)

v0.13.0

  • Conditional distributions and correlations for analysing the posterior (#321)
  • Moved rarely used arguments from pairplot into kwargs (#321)
  • Sampling from conditional posterior (#327)
  • Allow inference with multi-dimensional x when appropriate embedding is passed (#335)
  • Fixes a bug with clamp_and_warn not overriding num_atoms for SNRE and the warning message itself (#338)
  • Compatibility with Pyro 1.4.0 (#339)
  • Speed up posterior rejection sampling by introducing batch size (#340, #343)
  • Allow vectorized evaluation of numpy potentials (#341)
  • Adds vectorized version of numpy slice sampler which allows parallel log prob evaluations across all chains (#344)

v0.12.2

  • Bug fix for zero simulations in later rounds (#318)
  • Bug fix for sbi.utils.sbiutils.Standardize; mean and std are now registered in state dict (thanks @plcrodrigues, #325)
  • Tutorials on embedding_net and presimulated data (thanks @plcrodrigues, #314, #318)
  • FAQ entry for pickling error

v0.12.1

  • Bug fix for broken NSF (#310, thanks @tvwenger).

v0.12.0

  • Add FAQ (#293)
  • Fix bug in embedding_net when output dimension does not equal input dimension (#299)
  • Expose arguments of functions used to build custom networks (#299)
  • Implement non-atomic APT (#301)
  • Depend on pyknos 0.12 and nflows 0.12
  • Improve documentation (#302, #305, thanks to @agramfort)
  • Fix bug for 1D uniform priors (#307).

v0.11.2

  • Fixed pickling of SNRE by moving StandardizeInputs (#291)
  • Added check to ensure correct round number when presimulated data is provided
  • Subclassed Posterior depending on inference algorithm (#282, #285)
  • Pinned pyro to v1.3.1 as a temporary workaround (see #288)
  • Detaching weights for MCMC SIR init immediately to save memory (#292)

v0.11.1

  • Bug fix for log_prob() in SNRE (#280)

v0.11.0

  • Changed the API to do multi-round inference (#273)
  • Allow to continue inference (#273)

v0.10.2

  • Added missing type imports (#275)
  • Made compatible for Python 3.6 (#275)

v0.10.1

  • Added mcmc_parameters to init methods of inference methods (#270)
  • Fixed detaching of log_weights when using sir MCMC init (#270)
  • Fixed logging for SMC-ABC

v0.10.0

  • Added option to pass external data (#264)
  • Added setters for MCMC parameters (#267)
  • Added check for density_estimator argument (#263)
  • Fixed NeuralPosterior pickling error (#265)
  • Added code coverage reporting (#269)

v0.9.0

  • Added ABC methods (#250)
  • Added multiple chains for MCMC and new init strategy (#247)
  • Added options for z-scoring for all inference methods (#256)
  • Simplified swapping out neural networks (#256)
  • Improved tutorials
  • Fixed device keyword argument (#253)
  • Removed need for passing x-shapes (#259)

v0.8.0

  • First public version