Decorator Example in Data Science

python_decorator.md

Decorator Example in Data Science

This example explains Python decorators in the context of data science. The example acts as a quick reminder, rather than a complete guide.

Consider a Pandas DataFrame about posts on a social media. The DataFrame, called posts, contains a column with the number of likes for each post.

post_id	...	likes	...
1	...	43	...
2	...	92	...
3	...	54	...

The following function calculates the average number of likes per post.

def average_likes(data):
    return data['likes'].mean()

We would like to decorate the function with a decorator checking that the likes column has integer type. Such a decorated function may look like the following.

@data_column_has_int_type('likes')
def average_likes(data):
    return data['likes'].mean()

The decorator can be defined in the following way.

def data_column_has_int_type(column):
    def decorator(function):
        def wrapper(*args, **kwargs):
            data = args[0]
            if not pandas.api.types.is_integer_dtype(data[column]):
                raise ValueError(f"Column {column} does not have integer type.")
            return function(*args, **kwargs)
        return wrapper
    return decorator

The decorated function, average_likes, is equivalent to:

data_column_has_int_type('likes')(average_likes)(posts)

This composition of functions unwraps as:

• data_column_has_int_type('likes') ⟶ decorator with column set to 'likes', equivalent to:

  def decorator(function):
      def wrapper(*args, **kwargs):
          data = args[0]
          if not pandas.api.types.is_integer_dtype(data['likes']):  # <- See change
              raise ValueError(f"Column {column} does not have integer type.") 
          return function(*args, **kwargs)
      return wrapper

• decorator(average_likes) ⟶ wrapper with function set to 'average_likes', equivalent to:

  def wrapper(*args, **kwargs):
      data = args[0]
      if not pandas.api.types.is_integer_dtype(data['likes']):
          raise ValueError(f"Column {column} does not have integer type.")
      return average_likes(*args, **kwargs)  # <- See change

• wrapper(posts) becomes:

  if not pandas.api.types.is_integer_dtype(posts['likes']):
          raise ValueError(f"Column {column} does not have integer type.")
  average_likes(posts)  # <- See change

This flow can be viewed compactly as,

data_column_has_int_type('likes')(average_likes)(data)
                        decorator(average_likes)(data)  # column='likes'
                                         wrapper(data)  # column='likes', function=average_likes

https://artemrudenko.wordpress.com/2013/04/15/python-why-you-need-to-use-wraps-with-decorators/

@kurtu5, you are correct, and it is good to use @wraps in decorators. With @wraps, the undecorated function name and docstring would be preserved in the decorated function.

I omitted them from the short guide as I wanted to focus on the logic of decorators.

I leave the link to @wraps in the Python documentation here for those interested to learn more about it: https://docs.python.org/3/library/functools.html#functools.wraps