I have saved a keras model in some directories in dbfs to load and retrain that with more data, etc. The problem is that when cluster stops and restarts, seems those directories and model are no longer available there and it starts training a new model.

Any idea?