Lampe

💥 Breaking news

The required version of Zuko has been upgraded to 0.2.1 to follow API changes. Accordingly, the support for older PyTorch versions (1.11 and earlier) was dropped. PyTorch 1.12 has been around for about a year and is compatible with all CUDA drivers since 10.2.

✨ What's new

* Migrate the documentation to [lampe.readthedocs.io](https://lampe.readthedocs.io) (1cf58efe0223b642875c6bb307a85b552fba0220)
* Simplify HDF5 dataset (3b52b13af56d2e189b1d435fcc5c2f04305e0d17)
* Add in-memory dataset (f1a003a4f7fa743e1987a7b682661bd4b6ff1503)
* Add the contrastive neural ratio estimation (CNRE) inference algorithm (7) by bkmi
* Add the flow matching posterior estimation (FMPE) inference algorithm (12)

**Full Changelog**: https://github.com/probabilists/lampe/compare/v0.7.0...v0.8.0

✨ What's new

* Drop support for standardization with respect to the first and second moments (d6fdca7e73f43af89ef08dcdddb4a5b111c32b08)
* New neural score estimation (NSE) inference algorithm (7) with the help of glouppe

🐛 Bug fixes

* Fix incorrect grid shape in `utils.gridapply` for one-dimensional space (8)

**Full Changelog**: https://github.com/probabilists/lampe/compare/v0.6.0...v0.7.0

💥 Breaking news

In the [v0.4.2](https://github.com/probabilists/lampe/releases/tag/v0.4.2) release, LAMPE dropped the `nflows` package for built-in normalizing flows. Since then, the flow implementations have improved and new ones have been added, making the `lampe.nn.flows` module a respectable normalizing flow collection. In this context, we decided to export said module into a standalone package called [Zuko](https://github.com/probabilists/zuko). Doing so, we wish to dissociate the addition/development of simulation-based inference algorithms from the addition of flow architectures.

📝 Documentation

* Use `autosummary` for automatic API generation (822c3cc44292b1fc0b76e31bd537b37c63eae601)

**Full Changelog**: https://github.com/probabilists/lampe/compare/v0.5.5...v0.6.0

✨ What's new

* Introduce the `lampe.diagnostics` module (7ad196e3c685c0840d0de34aca2235d7502ba4d0)
* New expected coverage diagnostic (7ad196e3c685c0840d0de34aca2235d7502ba4d0)

**Full Changelog**: https://github.com/probabilists/lampe/compare/v0.5.2...v0.5.5

✨ What's new

* Implement unconstrained monotonic neural network (UMNN) based flows (14210bb293601dbdd210515afbb10427f4e96e19)
* Add option to overwrite HDF5 files (deda51efa9597cbdc2a62eb403cc778474ae5f7a)
* New balanced neural ratio estimation (BNRE) loss (3) by ADelau
* Simplify monotonic transforms (28ffeda251377c3b09fbb2b968a96e9d7e605de4)

**Full Changelog**: https://github.com/probabilists/lampe/compare/v0.5.0...v0.5.2

✨ What's new

* Drop `lampe.simulators` module (68caf4957e939b9643d0d4f8e564ebd0f6331b67)
* Reformat code with [Black](https://github.com/psf/black) (a676163badaeb1c7a67dd6ff2a275bd297360d3d)
* Enable random feature order in autoregressive flows (4858fd736f400da6ea631a83f1c3d65632d51b15)

🐛 Bug fixes

* Minor fixes (a3748039b0eef670a16da13526c3f45ffd64d37f)
* Add unit tests and fix detected bugs (5717651b7b7217e3a5275c2a469bb5184656167d)

📝 Documentation

* Add contributing guidelines (ebc60babcc7a62a26ed8a4b061f43aff1f91f0bf)

**Full Changelog**: https://github.com/probabilists/lampe/compare/v0.4.4...v0.5.0