AutoEncoderToolkit.jl - 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.

Scientific Software - Peer-reviewed - Julia
Published by mrazomej over 1 year ago

AutoEncoderToolkit.jl - 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.

Scientific Software - Peer-reviewed - Julia
Published by mrazomej over 1 year ago

AutoEncoderToolkit.jl - 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 julia struct JointGaussianLogDecoder <: AbstractGaussianLogDecoder decoder::Flux.Chain µ::Flux.Dense logσ::Flux.Dense end to julia 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.

Scientific Software - Peer-reviewed - Julia
Published by mrazomej over 1 year ago

AutoEncoderToolkit.jl - 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: https://github.com/mrazomej/AutoEncoderToolkit.jl/compare/v0.0.2...v0.0.3

Scientific Software - Peer-reviewed - Julia
Published by mrazomej almost 2 years ago

AutoEncoderToolkit.jl -

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

Scientific Software - Peer-reviewed - Julia
Published by mrazomej almost 2 years ago

AutoEncoderToolkit.jl - v0.0.1

AutoEncoderToolkit v0.0.1

Scientific Software - Peer-reviewed - Julia
Published by github-actions[bot] almost 2 years ago