My aim is to implement a class that inherits from pd.Series class and acts as a container object. This object will hold varied objects as its elements in the form of

container = Container(
   a = 12,
   b = [12, 10, 20],
   c = 'string',
   name='this value',
)

I will be accessing these elements via a dot notation:

print(container.a)     # Output: 12
print(container.b)     # Output: [12, 10, 20]
print(container.c)     # Output: string
print(container.name)  # Output: None

I have tried the following implementation:

import pandas as pd

class Container(pd.Series):
    def __init__(self, **kwargs):
        super().__init__(data=kwargs)

This works to a large extent.

However, there would be special cases. If index of one of the elements in the container is also a property in pd.Series() then the property would be returned. In the case of above example, container.name would return None value as it returns the pd.Series().name property of the series. I want container.name to return 'this value'.

So, maybe I need to overerite the __getattribute__ or __getattr__ to make this work.

I have tried the following:

def __getattr__(self, item):
    if item in self:
        return self[item]
    else:
        raise AttributeError(f"'{self.__class__.__name__}' object has no attribute '{item}'")

def __getattribute__(self, item):
    try:
        # Try to get an item using dot notation when it's available in the series itself
        value = super().__getattribute__('__getitem__')(item)
        if item in self:
            return value
    except (KeyError, AttributeError):
        pass
    # Fallback to the default __getattribute__
    return super().__getattribute__(item)

However, this would always return RecursionError: maximum recursion depth exceeded. I have tried different ways.

Note that I like to inherit from pd.Series() to have access to all other functionalities of the series.

My aim is to implement a class that inherits from pd.Series class and acts as a container object. This object will hold varied objects as its elements in the form of

container = Container(
   a = 12,
   b = [12, 10, 20],
   c = 'string',
   name='this value',
)

I will be accessing these elements via a dot notation:

print(container.a)     # Output: 12
print(container.b)     # Output: [12, 10, 20]
print(container.c)     # Output: string
print(container.name)  # Output: None

I have tried the following implementation:

import pandas as pd

class Container(pd.Series):
    def __init__(self, **kwargs):
        super().__init__(data=kwargs)

This works to a large extent.

However, there would be special cases. If index of one of the elements in the container is also a property in pd.Series() then the property would be returned. In the case of above example, container.name would return None value as it returns the pd.Series().name property of the series. I want container.name to return 'this value'.

So, maybe I need to overerite the __getattribute__ or __getattr__ to make this work.

I have tried the following:

def __getattr__(self, item):
    if item in self:
        return self[item]
    else:
        raise AttributeError(f"'{self.__class__.__name__}' object has no attribute '{item}'")

def __getattribute__(self, item):
    try:
        # Try to get an item using dot notation when it's available in the series itself
        value = super().__getattribute__('__getitem__')(item)
        if item in self:
            return value
    except (KeyError, AttributeError):
        pass
    # Fallback to the default __getattribute__
    return super().__getattribute__(item)

However, this would always return RecursionError: maximum recursion depth exceeded. I have tried different ways.

Note that I like to inherit from pd.Series() to have access to all other functionalities of the series.

• python
• pandas
• overriding

1 Answer 1

First, check if item exists in self using super().__contains__(item). If it exists, return self[item] directly. Otherwise, fall back to the default __getattribute__.

Here is the solution:

import pandas as pd

class Container(pd.Series):
    def __init__(self, **kwargs):
        object.__setattr__(self, "_custom_attrs", kwargs)
        super().__init__(kwargs)

    def __getattr__(self, item):
        if item in self._custom_attrs:
            return self._custom_attrs[item]
        if item in self.index:
            return super().__getitem__(item)
        raise AttributeError(f"'{self.__class__.__name__}' object has no attribute '{item}'")

    def __getattribute__(self, item):
        if item in {"_custom_attrs", "index", "dtype", "_mgr", "_data"}:
            return object.__getattribute__(self, item)
        
        custom_attrs = object.__getattribute__(self, "_custom_attrs")
        if item in custom_attrs:
            return custom_attrs[item]

        return pd.Series.__getattribute__(self, item)

container = Container(
    a=12,
    b=[12, 10, 20],
    c='string',
    name='this value'
)

print(container.a)     
print(container.b)     
print(container.c)     
print(container.name)