v1.0.0 DeepSaki - Major Release

Changes:

• Framework:

  • Upgrade from python 3.8 to python 3.10 since it supports a wide range of tensorflow version(2.6-2.14)
  • update of project structure to a modern SW project
  • automated CI/CD
  • docstrings and type annotations
  • initial automated documentation using mkdocs and mike supporting multiple version. Will be further improved in a future.
  • automated code coverage reports
  • automated publishing of packages on release
  • add changelog and contribution documentation
  • providing a dockerfile for the dev environment

• DeepSaki.initializers:

  • renamed DeepSaki.initializer -> DeepSaki.initializers
  • Refactored function make_initializer_complex() into a class called ComplexInitializer

• DeepSaki.layers:

  • renamed ResidualIdentityBlock -> ResidualBlock
  • New Layers:
    • LearnedPooling
    • FFT3D
    • iFFT3D

• DeepSaki.losses:

  • New abstract base class for image based losses.

• DeepSaki.optimizers:

  • renamed module from optimizer -> optimizers
  • new method switch_optimizer() to change the current optimizer of SwatsAdam or SwatsNadam optimizer.

• DeepSaki.augmentations:

  • renamed module from regularizations to augmentations since it only contained augmentations.

• DeepSaki.utils:

  • renamed DetectHw() -> detect_accelerator()
  • renamed EnableXlaAcceleration() -> enable_xla_acceleration()
  • renamed EnableMixedPrecision() -> enable_mixed_precision()

• New modules:

  • DeepSaki.types: provides type definitions and enums used in the package.

• Misc:

  • general code cleaning and refactoring.
  • changed print expressions to logging
  • using pyproject.toml for project config

• Outlook:

  • improved changelogs with issue tickets
  • new modules are planned, e.g. a diffusion module and a framework module featuring CycleGAN and DDPM.