save functionality - Githubissues

NLeSC / mcfly

A deep learning tool for time series classification and regression

Apache License 2.0

362 stars 82 forks source link

save functionality #63

Closed vincentvanhees closed 7 years ago

vincentvanhees commented 8 years ago

save model weights and architecture:

model.to_json()
model.save_weights() decide whether to save in mcfly as:
2 seperate files for weights and architecture
a folder containing the 2 files
a zip containing the 2 files (looks nice, but hides technology to the user)

The filename(s) will be user-defined

possibly with load functionality later

vincentvanhees commented 8 years ago

I implemented storage functionality to store the model as two seperate files (weights and architecture seperate) in a folder chosen by the user and a model-name chosen by the user. The model name and the folder are the identifiers the load function needs to retrieve the model. I choose for a custom implementation rather then relying on the Keras function, which I have reported in issue #67.

cwmeijer commented 8 years ago

There are no tests yet for this code right? You could define model and save it and load it again and then do some assertions on the loaded model. This would be good test for both functionalities.