Checking for Series Types (Representation Series, Tokenized Series, ...).

[henrifroese]

This is mainly meant as a first step towards closing #60 and should not be merged yet. It implements three types of custom pandas Series that implement the most common usages in the library. Also adds decorators for easier development using the types. See the discussion in #60 for more information.

[jbesomi]

Henri you are awesome!! 🎉🎉

[henrifroese]

We have now worked on this again; should be ready for a review/discussion (and maybe a merge) @jbesomi .

Summary (everything is explained more extensively in docstrings in _helper.py):

• Add Classes TextSeries, TokenSeries, VectorSeries, DocumentRepresentationSeries where each class has a function that can check for a Series s if s fulfills the type
• Add Decorator InputSeries(allowed_hero_series_type=) to check the input series type of a function
• Tests for the decorator

Use Cases:

We can now do two things:

>>> import pandas as pd
>>> from texthero._helper import *
>>> def tfidf(s: TokenSeries) -> DocumentRepresentationSeries:
>>>      # tfidf stuff

This is very expressive for the users in the documentation (and also for the developers)! Everyone can immediately see what tfidf is working on and what the output looks like. Doing help(TokenSeries) now also explains to the users what's going on.

This does not force users to do anything differently from before, they should just continue using Pandas Series as usual; it's mainly for documentation / ease of use / expressiveness. This also automatically adds good explainations through the docstrings.

Second of all, the decorator (probably self-explanatory):

>>> from texthero._helper import *
>>> import pandas as pd
>>> @InputSeries(TokenSeries)
... def f(s):
...     pass
>>> f(pd.Series("Not tokenized")) # doctest: +SKIP
>>> # throws a type error with a nice explaination
>>> f(pd.Series([["I", "am", "tokenized"]]))
>>> # passes

All of this is implemented very lightweight and adds basically no overhead; the type checking etc. is all O(1)

[jbesomi]

Amazing job! 🎉 That's soo nice and soo cool (and soo simple too) !!

[henrifroese]

Just renamed "DocumentRepresentationSeries" to "RepresentationSeries"

[jbesomi]

Everything looks great! I will merge it now ... the only observation: what if we move this under a separate file? Like types.py or heroseries.py or something like that?