Option share_hidden_layers does not work as expected

[tabergma]

Rasa version: current master
Python version: 3.7

Issue:
If the option share_hidden_layers is enabled in the DIETClassifier, we have a check which verifies that the feature dimension of the text and label features is the same. However, we also have an additional check that causes issues: That check compares the data_signature of the text and label features. data_signature stores the type of the feature and the shape of the first example. For example, it could be that the shape of text_features is 2x53 and the shape of label_features is 1x53. In this case the check fails, but it should not.

Error (including full traceback):

Traceback (most recent call last):
  File "/Users/tabergma/Repositories/rasa/rasa/__main__.py", line 119, in <module>
    main()
  File "/Users/tabergma/Repositories/rasa/rasa/__main__.py", line 108, in main
    cmdline_arguments.func(cmdline_arguments)
  File "/Users/tabergma/Repositories/rasa/rasa/cli/train.py", line 142, in train_nlu
    additional_arguments=extract_nlu_additional_arguments(args),
  File "/Users/tabergma/Repositories/rasa/rasa/train.py", line 431, in train_nlu
    additional_arguments,
  File "uvloop/loop.pyx", line 1456, in uvloop.loop.Loop.run_until_complete
  File "/Users/tabergma/Repositories/rasa/rasa/train.py", line 472, in _train_nlu_async
    additional_arguments=additional_arguments,
  File "/Users/tabergma/Repositories/rasa/rasa/train.py", line 506, in _train_nlu_with_validated_data
    **additional_arguments,
  File "/Users/tabergma/Repositories/rasa/rasa/nlu/train.py", line 90, in train
    interpreter = trainer.train(training_data, **kwargs)
  File "/Users/tabergma/Repositories/rasa/rasa/nlu/model.py", line 191, in train
    updates = component.train(working_data, self.config, **context)
  File "/Users/tabergma/Repositories/rasa/rasa/nlu/classifiers/diet_classifier.py", line 716, in train
    config=self.component_config,
  File "/Users/tabergma/Repositories/rasa/rasa/nlu/classifiers/diet_classifier.py", line 1032, in __init__
    self._check_data()
  File "/Users/tabergma/Repositories/rasa/rasa/nlu/classifiers/diet_classifier.py", line 1099, in _check_data
    "If hidden layer weights are shared, data signatures "
ValueError: If hidden layer weights are shared, data signatures for text_features and label_features must coincide.

Command or request that led to error:

rasa train

Content of configuration file (config.yml) (if relevant):

language: en
pipeline:
  - name: WhitespaceTokenizer
  - name: CountVectorsFeaturizer
    use_shared_vocab: True
  - name: DIETClassifier
    epochs: 100
    share_hidden_layers: True