✨ Highlights

• feat: addCategoricalMADE by@jnsbck in#1269(Major New Feature)
• tests:mini-sbibm by@manuelgloeckler in#1335(Major New Feature)
• feat: Score-based iid sampling by@manuelgloeckler in#1381(Major New Feature)
• Drop python3.9 support, fix ci by@janfb in#1412(Python Version Support Change)
• additional features for NPSE by@gmoss13 in#1370(Enhancement)

🐛 Bug Fixes

• fix: leakage correction breaks consistency of log prob vs log prob batched by@manuelgloeckler in#1355
• fix#1343 device handling in mog_log_prob by@janfb in#1356
• Fix failing tutorials, change MNLE default for log_transform to False by@janfb in#1367
• Fix conditional posterior shape and device bugs. by@janfb in#1373
• fix: type fix in VI subclasses. xfail pymc tests. by@janfb in#1390
• Temporary Wrappers to fix MADE by@gmoss13 in#1398
• Fix mnle tests, MCMC init pbar, sir batch size. by@janfb in#1410
• fix mnle tests by@janfb in#1415
• fix: protocol and refactor for custom potential by@janfb in#1409
• fix docs workflow by@janfb in#1419
• fix: gpu-handling for CategoricalMADE by@janfb in#1448

🛠️ Maintenance & Improvements

• increase tutorial test timeout by@janfb in#1360
• add nan check to _loss method. by@janfb in#1361
• Update and pin pre commit and ruff to recent version. by@janfb in#1358
• add nan handling in diagnostics by@janfb in#1359
• Improve tests to detect circular imports and resolve all of them by@manuelgloeckler in#1357
• docs: add details about zuko density estimators by@janfb in#1387
• fix: duplication of 'large number' in warning by@turnmanh in#1391
• improve PR and issue templates by@janfb in#1399
• perf: speed up CI with uv by@janfb in#1400
• tests: add pytest testmon plugin to speed up CI by@janfb in#1402
• small fixes to score methods by@janfb in#1404
• docs: Corrected python version in installation document (Revised) by@VijaySamant4368 in#1423
• docs: Markdown formatting compliant to Markdown Linter (solves#1434) by@nMaax in#1443

🎉 New Contributors

• @VijaySamant4368 made their first contribution in#1423
• @nMaax made their first contribution in#1443

Full Changelog:v0.23.3...v0.24.0

Contributors

v0.23.3

Highlights 🤩

• docs: Add conda-forge install instructions by@matthewfeickert in#1340
• feat:NLE with multiple iid conditions by@janfb in#1331

What's Changed 🚧

• fix: Correted typo in y-axis label by@turnmanh in#1296
• docs: update embedding networks notebook by@emmanuel-ferdman in#1297
• fix pickle issues in MCMC posterior + test by@manuelgloeckler in#1291
• Minor fix for EnsemblePosterior weights.setter by@CompiledAtBirth in#1299
• Remove deprecated neural_net access fromutils by@tvwenger in#1302
• [test] add tests for ensemble posterior weights by@samadpls in#1307
• Clarify last round behavior of SNPE-A by@michaeldeistler in#1323
• expose batched sampling option; error handling by@janfb in#1321
• Fix#1316: remove sample_dim docstring for condition. by@janfb in#1338
• docs: fix tutorial typos by@janfb in#1341
• docs: run and seed SBC tutorial by@manuel-morales-a in#1336

New Contributors 🎉

• @emmanuel-ferdman made their first contribution in#1297
• @CompiledAtBirth made their first contribution in#1299
• @tvwenger made their first contribution in#1302
• @matthewfeickert made their first contribution in#1340
• @manuel-morales-a made their first contribution in#1336

Full Changelog:v0.23.2...v0.23.3

Contributors

Bug Fixes

• fixup for failing hmc test by@michaeldeistler (#1247)
• fix: make RestrictedPrior a distribution to enable log_prob@janfb (#1257)
• fix: npe iid handling by@janfb (#1262)
• fix: tutorials test error handling, fix bugs in tutorials by@janfb (#1264)
• fix#1260: include points in plotting limits by@janfb (#1265)
• fix: conditioned potential error handling by@janfb,@michaeldeistler (#1275,#1289)
• fix: Allow 1D pytorch distributions by@michaeldeistler (#1286)

Documentation

• Rename SNPE to NPE in the README by@michaeldeistler (#1248)
• update pickling FAQ by@michaeldeistler (#1255)
• Adding example for custom DataLoader to tutorial 18 by@psteinb (#1256)
• docs: add readme intro to docs landing page by@janfb (#1272)
• Change sampling method for LC2ST tosample_batched() by@JuliaLinhart (#1279)

Maintenance

• Refactor simulate_for_sbi location by@samadpls (#1253)
• build: devcontainer update by@janfb (#1252)
• fix: docker notebook python version by@janfb (#1258)
• refactor: remove outputs except plots from tutorials. by@janfb (#1266)
• build: automatic nb stripping and pypi upload by@janfb (#1267)
• refactor: remove deprecated x_shape where not needed by@janfb (#1271)
• more explicit error message for CNN shapes by@Ankush7890 (#1281)

Contributors

• fix: includescore folder by adding__init__.py (#1245#1246)

Announcements

• Re-licensing: license change from
  AGPLv3 to
  Apache-2.0 (see#997 for
  details)
• sbi is now affiliated withNumFOCUS 🎉
• New contributors 🎉:@anastasiakrouglova,@theogruner,@felixp8,@Matthijspals,
  @jsvetter,@pfuhr,@turnmanh,@fariedabuzaid,@augustes,@zinaStef,@Baschdl,
  @danielmk,@lisahaxel,@janko-petkovic,@samadpls,@ThomasGesseyJonesPX,@schroedk

Major Changes

• internal renaming of all inference classes from, e.g.,SNPE toNPE (i.e., we
  removed theS prefix). The functionality of the classes remains the same. The NPE
  class handles both the amortized and sequential versions of neural posterior
  estimation. An alias for SNPE (and other sequential methods) still exists for
  backwards compatibility (#1238) (@michaeldeistler).
• changesbi default parameters:training_batch_size=200,num_chains=20 (#1221)
  (@janfb)
• change imports ofposterior_nn,likelihood_nn, andclassifier_nn. They should
  now be imported fromsbi.neural_nets, not fromsbi.utils (#994) (@famura)
• big refactoring of plotting utilities, new tutorial (#1084) (@Matthijspals)
• improved tutorials and website documentation (#1012,#1051,#1073) (@augustes,
  @zinaStef,@lisahaxel,@psteinb)
• improved website structure and contribution guides (#1019) (@tomMoral,@janfb)
• drop support for python3.8 and torch1.12 (#1233)
• refactor folder structure and naming ofneural_nets (#1237) (@michaeldeistler)

New Features

• full flexibility over the training loop (#983) (@michaeldeistler)
• unified density estimator classes (#952,#965,#979,#1151) (@michaeldeistler,
  @gmoss13,@tomMoral, @manualgloeckler)
• vectorized sampling and log_prob for(S)NPE given batches of x (#1153)
  (@manuelgloeckler,@michaeldeistler)
• batched sampling for vectorized MCMC samplers (#1176,#1210) (@gmoss13,@janfb)
• support@Zuko as a backend for normalizing flows (#1088,#1116)
  (@anastasiakrouglova)
• local c2st metric (#1109) (@JuliaLinhart)
• tarp coverage metric (#1106) (@psteinb)
• add interface for@pymc samplers (#1053) (@famura,@felixp8)
• flow matching density estimators (#1049) (@turnmanh,@fariedabuzaid,@janfb)
• score matching density estimators (#1015) (@rdgao,@jsvetter,@pfuhr,
  @manuelgloeckler,@michaeldeistler,@janfb)
• ABC methods for trial-based data using statistical distances (#1104)
  (@theogruner)
• support Apple MPS as gpu device (#912) (@janfb)
• dev container for usingsbi in codespaces on GitHub (#1070) (@turnmanh)
• enable importance sampling for likelihood-based estimators (#1183) (@manuelgloeckler)
• refactoring and unified shape handling forRatioEstimator (#1097) (@bkmi)
• faster sbc and tarp calibration checks via batched sampling (#1196) (@janfb)
• batched sampling and embedding net support forMNLE (#1203) (@janfb)
• adaptMNLE to new densitye stimator abstraction (#1089) (@coschroeder)
• better plotting options for coverage plots (#1039,#1212) (@janfb)
• allow for potential_fn to be a Callable (#943) (@michaeldeistler)

Bug Fixes

• bugfix for embedding net tutorial (#1159) (@Deismic)
• Fixup for process_x in EnsemblePosterior (#1148) (@Deismic)
• fixed notebook by changing MCMC parameters (#1058) (@zinaStef)
• fix: add NeuralPosteriorEnsemble to utils.init (#1002) (@jnsbck)
• fix: print_false_positive_rate (#976) (@danielmk)
• fix: make VIPosterior pickable (#951) (@manuelgloeckler)
• fix: bug in importance sampled posterior (#1081) (@max-dax)
• fix: embedding device and warning handling (#1186) (@janfb)
• fix: c2st with constant features (#1204) (@janfb)
• fix: erroneous warnings about different devices (#1225,@ThomasGesseyJonesPX)
• fix: type annotation in classConditionedPotential (#1222) (@schroedk)

Maintenance and other changes

• add pre-commit hooks (#955) (@janfb)
• add ruff to replaceisort,black,flake (#960,#978,#1113) (@janfb)
• switch topyproject.toml for package specification (#941) (@janfb)
• Split the GitHub workflow in CI and CD (#1063) (@famura)
• split linting process from the CI/CD workflow (#1164) (@tomMoral)
• Switch to the newestpyright and fix all typing errors (#1045,#1108) (@Baschdl)
• introduce two docs versions:latest pointing to latest release at
  https://sbi-dev.github.io/sbi/latest/ anddev pointing to the latest version on
  mainhttps://sbi-dev.github.io/sbi/dev/

Contributors

API change

• We have movedsbi to an new github organization:https://github.com/sbi-dev/sbi
• We have changed the website of thesbi docs:https://sbi-dev.github.io/sbi/.
• sbi.analysis.pairplot:upper was replaced byoffdiag and will be deprecated in a future release.

Features and enhancements

• size-invariant embedding nets for amortized inference with iid-data (@janfb,#808)
• option for new using MAF with rational quadratic splines (thanks to@ImahnShekhzadeh,#819)
• improved docstring forprocess_prior (thanks to@musoke,#813)
• extended tutorial for SBI with iid data (@janfb,#857)
• new tutorial for SBI with experimental conditions and mixed data (@janfb,#829)
• New options forpairplot:
  • upper is now calledoffdiag to match other kwargs.
  • alternating colors forsamples andpoints
  • option to add alegend and passkwargs for the legend.

Bug fixes

• fixed memory leak in inappend_simulations (thanks to @VictorSven,#803)
• bug fix for CNRE (thanks to@bkmi,#815)
• bug fix for iid-inference with posterior ensembles (@janfb,#826)
• bug fix for simulation-based calibration with VI posteriors (@janfb,#834,#838)
• bug fix for BoxUniform device handling (@janfb,#854,#856)
• bug fix for MAP estimates with independent priors (@janfb,#867)
• bug fix for tutorial on SBC (@michaeldeistler,#891)
• fix spurious seeding forsimulate_for_sbi (@jan-matthis,#876)
• bump python version of github action tests to3.9.13 (@michaeldeistler,#888,#900)

Contributors

• implementation of"Contrastive Neural Ratio Estimation" (thanks to@bkmi,#787)
• implementation of"Balanced Neural Ratio Estimation" (thanks to@ADelau,#779)
• bugfixes for SBC, device handling and handling of iid-observations (#793,#789,#780)

Contributors

Major changes and bug fixes

• implementation of"Truncated proposals for scalable and hassle-free sbi" (#754)
• sample-based expected coverage tests (#754)
• permutation invariant embedding to allow iid data in SNPE (thanks@coschroeder,#751)
• convolutional neural network embedding (thanks@coschroeder,#745,#751,#769)
• disallow invalid simulations when using SNLE, SNRE, or atomic SNPE-C (#768)

Enhancements

• add tutorial on all available methods (#754)
• allow seeding ofsimulate_for_sbi on multiple workers (#762)
• exposeenable_transforms in sampler interface (#756)
• bugfix for building the transformation of transformed distributions (#756)

Contributors

• Rely on new version ofpyknos with bugfix for APT with MDNs (#734)
• bugfix: atomic SNPE-C now allows any kind of proposal (#732)
• bugfix for SNPE with implicit prior on GPU (#730)
• SNPE-A hasforce_first_round_loss=True as default (#729)

• bug fix forArviZ integration (#727)