In this PR I'm removing binaries (H5 file) from files exported while training Sonnia model. Instead, I'm saving model configuration and weights in json format.
Model is loaded by initiating the object with the use of architecture given in model_overview.yaml file and by setting weights stored in model.json file. I decided to take this approach since Sonnia is using additional custom layers, which Keras struggles to unpack.
In this PR I'm removing binaries (H5 file) from files exported while training Sonnia model. Instead, I'm saving model configuration and weights in json format.
Sonnia is based on Keras Functional model, which doesn't support yaml (ref: https://www.tensorflow.org/api_docs/python/tf/keras/models/model_from_yaml). Thus, I decided to go with json format.
Model is loaded by initiating the object with the use of architecture given in model_overview.yaml file and by setting weights stored in model.json file. I decided to take this approach since Sonnia is using additional custom layers, which Keras struggles to unpack.