Releases: mathisgerdes/bijx
v1.2.1
- Improved API consistency in accepting and passing
**kwargsto constituent densities and transformations. - Fixed minor sign error in Fourier scaling layer (constant offset in log-density; irrelevant for training).
- Bump minimum python version as required by flax.
Full Changelog: v1.2.0...v1.2.1
v1.2.0
Add analytic bijections and radial flows, including short tutorial.
Full Changelog: v1.1.3...v1.2.0
v1.1.3
Add SDE solve to diffrax wrapper and small bug fix.
Full Changelog: v1.1.2...v1.1.3
v1.1.2
Release for JOSS paper. Corrected minor typos.
Full Changelog: v1.1.1...v1.1.2
v1.1.1
Update to latest flax version:
- Deprecates
.valueonnnx.Variable - New
nnx.dataclass
Full Changelog: v1.1.0...v1.1.1
v1.1.0
Crouch-Grossmann Integration API Refactor
- Make it easier to select various choices in the integration algorithm (e.g. enabling projection onto tangent space)
- Enhanced code clarity and future extensibility
Flax Framework Migration
- Migrated from flax.struct.dataclass to newer nnx.Pytree API
Minor improvements to batch handling and bug fixes.
Full Changelog: v1.0.2...v1.1.0
v1.0.2
- Bug fixes (primarily for GMM)
- Fix various compatibility issues with newest flax version.
Full Changelog: v1.0.1...v1.0.2
v1.0.1
- Added general multivariate and diagonal gaussian distributions.
- Added general mixture of (the same kind of) distributions.
- Minor improvements to documentation and added GMM tutorial.
Full Changelog: v1.0.0...v1.0.1
Initial stable release
JAX-based normalizing flows and bijections library, with special focus on physics research. Features continuous normalizing flows, matrix Lie group operations, and flexible architecture blocks. Built on Flax NNX with focus on research flexibility and composability.