v0.1.2 | JOSS publication

Release: AutoEncoderToolkit.jl v0.1.2

We’re thrilled to announce the release of AutoEncoderToolkit.jl version v0.1.2!

Highlights:

•	Type stability: All custom structs are now type-stable, possibly enhancing the performance.
•	Bug Fixes: All bugs that made tests fail on `GitHub` have been fixed.
•	Updated Documentation: The documentation now includes community guidelines.

We especially thank the superb JOSS reviewers for their excellent suggestions and constructive feedback. The entire review process has been a remarkable experience that gives us hope that there is a way to fix the peer-review system, and JOSS has taken a significant step forward in the right direction.

v0.1.1 | All tests pass.

This version contains minor changes to facilitate testing with GitHub actions. Before, tests only passed locally. Now, replacing all Dict with NamedTuple, tests pass on GitHub.

v0.1.0 | CUDA compatibility fully tested

Changes

1. Fully tests all CUDA-related functionality. This includes

   • Custom GPU kernels
   • Training functions for all available models

2. Encoders and Decoders are now type stable. For example JointGaussianLogDecoder went from

 struct JointGaussianLogDecoder <: AbstractGaussianLogDecoder 
     decoder::Flux.Chain 
     µ::Flux.Dense 
     logσ::Flux.Dense 
 end

to

 struct JointGaussianLogDecoder{D,M,L} <: AbstractGaussianLogDecoder 
     decoder::D
     µ::M
     logσ::L
 end

However, because of the depth of these nested structs, the function typeof only returns JointGaussianLogDecoder{...}, hiding all extra subtypes.

3. Custom layers now use Flux.@layer rather than Flux.@functor.

Fixing bugs when training RHVAEs on GPU

This release fixes some bugs on the CUDA extension to be able to train RHVAEs on CUDA GPUs

Full Changelog: v0.0.2...v0.0.3

v0.0.2

This is a minor fix that changes the computation of the log-likelihood function for the BernoulliDecoder and the CategoricalDecoder.

v0.0.1

AutoEncoderToolkit v0.0.1