use pre-trained model in predictor

[zeke]

Weaning ourselves off the notebook. We were training the classifier every time we imported the notebook. Now we're pre-training the classifier and importing it into the predictor.

🍐 @andreasjansson

[zeke]

Some extra notes from our pairing session:

• The notebook is an interactive aid for initially training the model
• If we were doing training frequently, we would start extracting the functions into standalone python modules
• importing a whole notebook is an anti-pattern
• imports should be immediate. No I/O should happen at import time other than reading the python files