Fix several bugs, prepare PyTorch hub support

[movabo]

Bugs fixed:

• AttributeError: spacy.pipeline.DependencyParser object has no attribute 'labels' -> Allow spacy versions greater 2.0.*
• ModuleNotFoundError: No module named 'models.singletarget.stancedetectionpaper' -> Local import in method where it's required
• ModuleNotFoundError: No module named 'models.singletarget.knowledgesourceshosseiniatdbert' -> Local import in method where it's required

[movabo]

The PR is ready for merge now. When you want add a new version of the model, you can update this by adding

@provide_pretrained("vX.Y.Z", "https://github.com/fhamborg/NewsMTSC/releases/download/vX.Y.Z/grutsc")

to grutscsingle.py. To automatically download this version when using --state_dict pretrained, adjust the following lines:

- @default_pretrained("v1.0.0")
+ @default_pretrained("vX.Y.Z")

If you want to add pretrained models for other architectures (e.g. LCF_BERT), change following lines:

• /models/signletarget/lcf.py:

  - from models.FXBaseModel import FXBaseModel
  + from models.FXBaseModel import FXBaseModel, provide_pretrained, default_pretrained
  +
  + @default_pretrained("vX.Y.Z")
  + @provide_pretrained("vX.Y.Z", "https://github.com/fhamborg/NewsMTSC/releases/download/vX.Y.Z/<NAME_OF_LCF_BERT>")
  class LCF_BERT(FXBaseModel):

• hubconf.py

  + from models.singletarget.lcf import LCF_BERT

  Then, the newly added model should be automatically downloaded when e.g. executing python infer.py --own_model_name lcf_bert.

[fhamborg]

I'm sorry, it seems that my notes weren't submitted yet (wrote them 1-2 days ago, but didnt hit the submit button)

[movabo]

I implemented now a download utility as per my last comment. For a simple download to get NewsMTSC running, python download.py is sufficient.

Also, I added version handling to the download utility. Help-text:

usage: download.py [-h] [--own_model_name OWN_MODEL_NAME] [--version VERSION]
                   [--force] [--list_versions]

optional arguments:
  -h, --help            show this help message and exit
  --own_model_name OWN_MODEL_NAME
  --version VERSION     version of the model to download, use --force to
                        overwrite a version which was already downloaded
  --force               force the download of a model and overwrite potential
                        previous versions
  --list_versions       List all pretrained model versions which a model
                        provides