Support dropping a column trough a transformer

[pvk-developer]

Problem Description

Currently our transformers are expected to always return a transformed column, but in some cases we would like to learn something from the column, drop it and then populate it with values during the reverse_transform.

Expected behavior

class PIITransformer(BaseTransformer):
    OUTPUT_TYPES = {
        'pii': 'categorical'
    }
    def __init__(self):
        self.length = None
        self.faker = Faker()
    
    def _fit(self, columns_data):
        self.length = len(columns_data)
        
    def _transform(self, columns_data):
        return None  # Drop the column
    
    def _reverse_transform(self, columns_data):
        return [self.faker.name() for i in range(self.length)]  # Generate the values

Which should lead to the following outcomes:

data = pd.DataFrame({
    'name': ['John', 'Doe', 'John Doe', 'John Doe John'],
    'donation': [20, 30, 40, 50]
})

pii_transformer = PIITransformer()
transformed = pii_transformer.fit_transform(data, 'name')
transformed

   donation
0        20
1        30
2        40
3        50

Which would contain only the donation value (the name has been dropped).

When reverse_transforme:

data = pii_transformer.reverse_transform(transformed)
data

   donation              name
0        20    Gregory Rivera
1        30  Melissa Peterson
2        40   Mrs. Lori Pitts
3        50       Susan Payne

A column name has been returned with random values, the data contains both name and donation columns.