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 SMCABC 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)