Creating attributes of attributes - possible?

Question

I am dealing with monthly data for a number of different data sets (e.g. air temperature, ocean temperature, wind speeds etc), whereby each month and each data set will share similar attributes. I want to be able to initialise and populate these attributes as efficiently as possible. So far I can only think of trying to somehow append new attributes to pre-existing attributes. Is this possible in Python?

For example, let's say I initialise all my monthly variables like the following:

class Data_Read:
    def __init__(self, monthly=np.zeros((200,200,40))): #3D data
        self.jan = monthly 
        self.feb = monthly
        self.march = monthly
        self.april = monthly
        self.may = monthly
        self.june = monthly
        self.july = monthly
        self.august = monthly
        self.sept = monthly
        self.oct = monthly
        self.nov = monthly
        self.dec = monthly

Then I can create a new data set for each month which will become something like air_temp.jan or wind_speed.june by doing the following:

air_temp = Data_Read()
wind_speed = Data_Read()

These would be the raw data attributes. However I would then like to do some processing to each of these data sets (like de-trending). Is there a way I can create a class (or a new class?) that will generate new attributes like self.jan.detrend. Basically I want to avoid having to write 12 lines of code for every attribute I want to create, and then be able to easily call any "attribute of the attribute" in my data processing functions thereafter.

Many thanks.

In your code above every month will get the *same* object and if you do not provide `monthly` even the instances will have the same object (the one from the default). — Klaus D., Mar 22 '19 at 16:35
Perhaps take a shape as input instead of an array, then create a new array with that shape for each month? Regarding your main question, you could just write methods like `detrend` on your `Data_Read` class. `detrend` can take a month name as input, so you could do `air_temp.detrend("jan")` to get that result. If you want to cache these results so that you don't recompute them each time, you could keep a dictionary or such of computed results internally. — Nathan, Mar 22 '19 at 16:38
You shouldn't have 12 separate variables in the first place. You want *one* attribute, a `dict` with months as keys. — chepner, Mar 22 '19 at 16:41
Thanks all. I had completely overlooked Dictionaries. I think these are the right way to go. — Will Gregory, Mar 22 '19 at 16:47
Though if you still want to be able to access the arrays as `air_temp.jan` when you internally have a dictionary, you can override `__getattr__` to look up month names in your dictionary. — Nathan, Mar 22 '19 at 16:49

score 0 · Answer 1 · answered Mar 22 '19 at 17:00

Here's an example of how you could store everything internally in a dictionary and still reference the arrays as attributes as well as call functions on arrays by name:

import numpy as np


class Data_Read:
    def __init__(self, monthly_shape=(200, 200, 40)): #3D data
        months = [
            "jan",
            "feb",
            "march",
            "april",
            "may",
            "june",
            "july",
            "august",
            "sept",
            "oct",
            "nov",
            "dec"
        ]
        self._months = {month: np.zeros(monthly_shape) for month in months}

    def detrend(self, month):
        # this is a dummy function that just increments
        return self._months[month] + 1

    def __getattr__(self, name):
        if name in self._months:
            return self._months[name]
        return super().__getattr__(name)

air_temp = Data_Read()
print(air_temp.jan.shape)  # (200, 200, 40)
print((air_temp.detrend("jan") == 1).all())  # True

You can also achieve the same result using setattr and getattr because attributes are just stored in a dictionary on the object anyway:

import numpy as np


class Data_Read:
    def __init__(self, monthly_shape=(200, 200, 40)): #3D data
        months = [
            "jan",
            "feb",
            "march",
            "april",
            "may",
            "june",
            "july",
            "august",
            "sept",
            "oct",
            "nov",
            "dec"
        ]
        for month in months:
            setattr(self, month, np.zeros(monthly_shape))

    def detrend(self, month):
        # this is a dummy function that just increments
        return getattr(self, month) + 1

Creating attributes of attributes - possible?

1 Answers1

Linked