Want to have two yaml file templates for the current best model. One template should be for training from scratch (and so should specify the model's architecture) and one - for continuing to train from a loaded .pkl file. In both templates the following should be configurable through the .json file: batch size, learning rate schedule, momentum, dropouts.
Want to have two yaml file templates for the current best model. One template should be for training from scratch (and so should specify the model's architecture) and one - for continuing to train from a loaded .pkl file. In both templates the following should be configurable through the .json file: batch size, learning rate schedule, momentum, dropouts.