Pass explict filenames, not subsets, for training

[JimCircadian]

• IceNet version: 0.2.9

Description

Running an icenet_train command, you don't pass the filename but rather the identifier for the dataset. This seems counter intuitive and I'm not entirely sure what it offers, ideally we should pass a filename. Thoughts @bnubald?

What I Did

$ icenet_train_tensorflow -nw -e 1 dataset_config.miniscule_bas_south.json local_test1 42

  File "/home/jambyr/icenet2/icenet/icenet/data/dataset.py", line 77, in __init__
    self._load_configuration(configuration_path)
  File "/home/jambyr/icenet2/icenet/icenet/data/dataset.py", line 130, in _load_configuration
    raise OSError("{} not found".format(path))
OSError: dataset_config.dataset_config.miniscule_bas_south.json.json not found

Note I'm using a dev branch so line numbers are awry, but this applies to the 0.2 release branches. We should change this and the equivalent of MergedIceNetDataSet

dataset = IceNetDataSet("dataset_config.{}.json".format(args.dataset),
                                batch_size=args.batch_size,
                                shuffling=args.shuffle_train)

[JimCircadian]

Will require the pipeline changing as well. Leaving here as a record for someone to address, might be worth checking for other instances of this type of behaviour.

[bnubald]

@JimCircadian, yes, is counter-intuitive, but assume would like to leave legacy option?

E.g. detect if file extension is json, and if it isn't, use existing approach, else, use filename.

[JimCircadian]

I like your style @bnubald, I'll stick this in to my existing work stream 😄