Fixed similarity_by_path returning None value.

[cyrilou242]

This pull requests aims to fix the issue https://github.com/alvations/pywsd/issues/31 described by @michael-aloys. (and more marginally https://github.com/alvations/pywsd/issues/41)

• Fixed typo in the max of the two path similarity The path_similarity function in wordnet is not commmutative.
  As said by @michael-aloys, there was a typo in max(wn.path_similarity(sense1,sense2), wn.path_similarity(sense1,sense2)) The 2 arguments in the max were the same, it is corrected to max(wn.path_similarity(sense1,sense2),wn.path_similarity(sense2,sense1))

• Fixed None return When no path between two senses is found, path_similarity and wup_similarity return None. This is the correct behavior of ntlk.wordnet and should not be changed. This is a problem because:

  • when None is returned by path_similarity, it directly breaks the python max described above (in python3)
  • when None is returned by wup_similarity, it can break the python max operation performed in the pywsd max_similarity function.

In this fix, checks on None value are added. If such case happen, the function now returns no_path_value, which is by default 0. This 0 value follows the behavior that was already implemented in the Leacock-Chodorow similarity case when the similarity was not computable.

To discuss:

I put no_path_value in argument instead of hardcoding 0 because i thought that in some cases one may want this value not to be 0. For instance: to avoid zeroing in a multiplication or dividing by 0 (by setting it to a very small float), or to put it to None to catch it after. Also, the default value 0 of no_path_value may be discussed.

It is important to notice the behavior I implemented for the path_similarity case: Sometimes wn.path_similarity(sense1,sense2) would return a value and wn.path_similarity(sense2,sense1) would return None. If the case happen of one path_similarity returning a number and the other returning None, the path_similarity that gave a number will be kept and returned

[alvations]

Resolve now that we have more controls over how wordnet is access through https://github.com/alvations/wordnet