[NLP] Rework model_rules.py

[LexABzH]

Problem

Model Rules was created to perform 'predictions' with simple human rules. However this model did not get any update for quite a moment and needs to be updated to the same level of quality as the other models. Plus, it is not tested.

Concerned template

• [x] NLP template
• [ ] NUM template
• [ ] VISION template
• [ ] How templates are generated - Jinja

Solution

• MVP: An updated class with new unit tests.

• Possible additions: Add e2e tests

[LexABzH]

Model rules should be functions, but we can't find a reliable way to save them. If they are model's methods, we can pickle an instance but it won't protect it from changes in our code.

For example :

• I create an instance TOTO with ruleA (let's say "OK" if len(x) > 10 else "KO")
• I save my instance TOTO as a pickle file (with pickle, dill, or w/e)
• I change method ruleA (let's say "OK" if len(x) > 25 else "KO")
• I reload my instance TOTO and try to get a new prediction. It will use the updated ruleA method instead of the old one.

This is not acceptable has we want the model to have the same results whenever we reload it.

We thought about using sklearn's FunctionTransformer, but it has the same behavior (see https://github.com/scikit-learn/scikit-learn/issues/12903 and https://github.com/scikit-learn/scikit-learn/issues/12904)

[LexABzH]

Note : it works fine with lambda functions