apmoore1 / bella Goto Github PK

Hi,
Thanks for starting this project, much needed and very useful.

Could you please make host & port of tweeboparser and corenlp configurable?

I had too many containers running and 9000 was already being used, so I had to change port in stanford_tools.py. Since pip install is being used for module installation, it would be better to provide configurable options so as to avoid need for changing source code.
Also, you might want to choose a different default port other than 9000, Since portainer which many people use for managing containers also like corenlp uses 9000 as it's default port.

When getting the configuration information for the parsers in a multiprocessing fashion, I believe the following error occurs when one thread has started to write and the other is reading thus the one that is reading sees a blank configuration files as it has been overwritten by the writing thread.

Error:

tweebo_api = TweeboParser()
  File "/home/andrew/Documents/Bella/bella/dependency_parsers.py", line 50, in __new__
    hostname, port = cls.get_config()
  File "/home/andrew/Documents/Bella/bella/dependency_parsers.py", line 34, in get_config
    if 'tweebo_parser' in config_data:
TypeError: argument of type 'NoneType' is not iterable
"""

The above exception was the direct cause of the following exception:

TypeError                                 Traceb

The Moses tokeniser does not work correctly when using it in a multiprocessing setup the issues this causes are the following:

1. The output of the multiprocessing setup when using map is not the same as the input
2. Causes a broken pipe

Make the machine learning models have the same interface or require the same mixin so that they contain the same basic methods with the same basic parameters e.g.:
fit(x, y, *args, **kwargs)
predict(x)

This will ensure the models are a lot easier to use and understand.

Large project would be to move out the Tensorflow models from this project into Bella-TF so that this project can remove several large dependencies:

1. tensorflow
2. Keras
3. pydot
4. graphviz