Fix module search path

[annafil]

👋 I found that when trying to run the given instructions, I was getting errors about not being able to find the text_recognizer module. Changing the instructions to invoke the emnist_dataset as part of a package triggers the right module search path and seems to do the trick (see updated readme below).

However, that then led to a runtime warning:

/usr/lib/python3.6/runpy.py:125: RuntimeWarning: 'text_recognizer.datasets.emnist_dataset' found in sys.modules aft
er import of package 'text_recognizer.datasets', but prior to execution of 'text_recognizer.datasets.emnist_dataset
'; this may result in unpredictable behaviour
  warn(RuntimeWarning(msg))

Commenting out the explicit call to .emnist_dataset in datasets/__init__.py seems to solve that. There's probably a reason that was in there in the first place though 👀 😬

Similarly, if ☝️ makes sense, the paths in https://github.com/full-stack-deep-learning/fsdl-text-recognizer-project/tree/master/lab2_sln/readme.md probably need adjusting as well. E.g.

pipenv run python3 -m training.run_experiment '{"dataset": "EmnistD
ataset", "dataset_args": {"subsample_fraction": 0.1}, "model": "CharacterModel", "network": "mlp"}

[switchfootsid]

Thanks! This solved a lot of my problems. Must be merged.