是否可以在不定义类的情况下以“ self”作为参数编写函数？

Question

I am trying to cut some lines of code by implementing functions. I was looking into functions that take themselves as an argument, as in oop. For example:

def drop_columns(self, columns):
    return self.drop(columns, axis = 1)

df.drop_columns(['id', 'date'])

or

def print_shape(self):
    print(self.shape)

df.print_shape()

But more sophisticated of course. However, the code does not work. If I feed it a data frame, it throws an error: AttributeError: 'DataFrame' object has no attribute 'print_shape'

Is there a way to make this work?

Answer 1

When you call a method on an instance of a class, that instance passes itself as the first parameter. We usually call that parameter self , for convenience.

This only works if the method is bound to the instance in the first place. When you do df.print_shape() , it doesn't work, because you never attached print_shape() to df in any way.

Both of the following will work, though:

# approach 1: call it as a function
print_shape(df)

# approach 2: assign the function as an attribute of df, then call it as a method
setattr(df, 'print_shape', print_shape)
df.print_shape()

Approach 1 is preferred, as it's generally better practice not to modify objects that you didn't create yourself (and not to do so dynamically like this). But knowing that Approach 2 exists gives some perspective/insight into how python works as a language. If you were sitting inside print_shape() , unable to look at anything going on outside of it, you wouldn't be able to tell the difference between the two approaches.

Answer 2

I think you are confusing classes and OOP with functions. In your case, don't treat your inputs as objects, rather arguments to your functions:

drop_columns(df, ['id', 'date'])
print_shape(df)

These calls should work.

Answer 3

Another approach here would be to subclass from DataFrame and build your convenience functions in this class:

import pandas as pd
import numpy as np

class EnhancedDataFrame(pd.DataFrame):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
    def print_shape(self):
        print(self.shape)
    def print_foo(self):
        print('This is the \'foo\'')

data = np.array(np.random.randint(0,100,(3,4)))
columns = ['A','B','A','C']
index = ['Row 1', 'Row 2', 'Row 3']
frame = EnhancedDataFrame(data, index=index, columns=columns)

And then you should be able to do like:

So getting to the root of your problem:

However, the code does not work. If I feed it a data frame, it throws an error: AttributeError: 'DataFrame' object has no attribute 'print_shape'

We've now implemented our own class ( EnhancedDataFrame ) which is a DataFrame (well, ok, technically it inherits from DataFrame ). It implements all of the methods you'd normally expect in a DataFrame , but also now includes whatever convenience methods you might want to add!

Answer 4

Taking the second example... What you can do is:

    def __init__(self, shape=None):
        self.shape = shape

Now, call print_shape(Shape()) , or print_shape(Shape('circle')) .

Since, print shape is not bound to any class (or object), it can treat self as a parameter.