0.2.5 - Better support for converter Keras models

PyTorch to Keras adds Activation layers after dense and an input layer out front.

0.2.4.1- Fix PyPi GH Action

bump

0.2.4 - Correctly display when user tries to convert unsupported layer

Fixes an issue when a model contains an unknown or unsupported layer, would show AttributeError as the function that looks up the layer returns False, when we expected an Unsupported() class to be returned.

Reported by Vasileios Kokorakis!

0.2.3 - Fix TF 2.3 activation error

• Fixes AttributeError: 'function' object has no attribute '_keras_api_names' when using tanh with TF 2.3

0.2.2 - Support LeakyReLU activation

• Support the LeakyReLU activation with custom alpha

0.2.1 Patch

Specify numpy version ~1.18.3 to avoid deprecation warning when running np.array() on non-array weights and biases. numpy 1.19 will work but will display deprecated bahavior warning. Further numpy releases may not work with Konverter when konverting

Release 0.2 - The Interface Update

• New custom CLI with pretty colors and emoji
  • Add support for a few different flags with CLI: custom indentation, whether Konverter should be silent, and whether to prepend watermark
• If an output file path is not given, use the input model's name and current directory to save konverted output model.

Bug fixes and minor changes:

• Fixed numpy deprecated behavior of creating array of objects without specifying dtype=object
• Removed click and typer dependencies

0.1.7 Patch

Fixes:

• 'Unsupported' object has no attribute 'info'
• Support differently-named Sequential models

0.1.6 Release

Support for BatchNormalization added. Works with all supported layers, along with Dropout

0.1.5 Official Release

• Update README.md with instructions on how to install and use.
• Add example files in examples/
• Fix tests