Set keras model.fit verbosity to 2

[aamster]

Currently deepinterpolation runs model.fit with the default verbosity of "auto". This does logging of training status in interactive mode, which uses tqdm. It clutters the logs and is not useful for production in which logs are logged to a file.

Current log looks like this

INFO:__main__:#0151/2 [==============>...............] - ETA: 10s - loss: 0.9100#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#0152/2 [==============================] - ETA: 0s - loss: 0.9100 #010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#010#0152/2 [==============================] - 14s 3s/step - loss: 0.9100

Set verbosity for the cli to 2, which will log on individual lines. The CLI will always be run in a non-interactive way.