`ZeroDivisionError` in determine_intent when tags are empty

[PLNech]

• I'm using an IntentDeterminationEngine along the lines of the Multi Intent parser Example, with no required entity type and several optional ones.

Stacktrace

  File "/home/pln/.PyCharm2019.3/config/scratches/scratch_12.py", line 16, in <module>
    print([i for i in engine.determine_intent(utterance, include_tags=False)])
  File "/home/pln/.PyCharm2019.3/config/scratches/scratch_12.py", line 16, in <listcomp>
    print([i for i in engine.determine_intent(utterance, include_tags=False)])
  File "/home/pln/.virtualenvs/nluservice/src/adapt-parser/adapt/engine.py", line 131, in determine_intent
    best_intent, tags = self.__best_intent(result, remaining_context)
  File "/home/pln/.virtualenvs/nluservice/src/adapt-parser/adapt/engine.py", line 81, in __best_intent
    i, tags = intent.validate_with_tags(parse_result.get('tags') + context_as_entities, parse_result.get('confidence'))
  File "/home/pln/.virtualenvs/nluservice/src/adapt-parser/adapt/intent.py", line 190, in validate_with_tags
    total_confidence = intent_confidence / len(tags) * confidence
ZeroDivisionError: float division by zero

Steps to replicate the Issue

1. Define an IntentDeterminationEngine as such:

   
   engine: IntentDeterminationEngine = IntentDeterminationEngine()
   engine.register_entity("Kevin", "who") # same problem if several entities
   builder: IntentBuilder = IntentBuilder("Buddies")
   builder.optionally("who") # same problem if several entity types
   engine.register_intent_parser(builder.build())

2. Use it on an utterance that contains an entity, get expected results

   print([i for i in engine.determine_intent("Kevin is a friend")]) # works

3. Use it on an utterance that contains no entity, get an exception

   print([i for i in engine.determine_intent("Julien is a friend")]) # raises

Can also be reproduced with an empty utterance "".

Be as specific as possible about the expected condition, and the deviation from expected condition.

I expect the engine to return an empty determination ([]) when no entity is recognized, instead of raising a generic ZeroDivisionError exception.

[PLNech]

I believe fixing this could just mean refactoring validate_with_tags. For now I'd see this as:

• Update the line 190 that divides by the length of a potentially empty collection: https://github.com/MycroftAI/adapt/blob/402d2d82048e5f7de2d8edd1c2efd9c9260f74bb/adapt/intent.py#L190

  • [ ] Could be wrapped in a conditional, e.g. total_confidence = [..] or 0 if not tags

• Take the opportunity to clarify the role of the confidence parameter, documented as a float: https://github.com/MycroftAI/adapt/blob/402d2d82048e5f7de2d8edd1c2efd9c9260f74bb/adapt/intent.py#L137-L142

  From my understanding, its value is always ignored, being redefined here: https://github.com/MycroftAI/adapt/blob/402d2d82048e5f7de2d8edd1c2efd9c9260f74bb/adapt/intent.py#L155

  Its final value is used to compute the total confidence in the line 190 I linked above. Thus, the new value of confidence defined line 192 is never used: https://github.com/MycroftAI/adapt/blob/402d2d82048e5f7de2d8edd1c2efd9c9260f74bb/adapt/intent.py#L192

  • [ ] Could remove confidence from validate_with_tags's parameters as it is never used
  • [ ] Could replace confidence by _ in line 192 as it is never used