在复制之后但在编辑复制之前编辑原始DataFrame会更改复制

Question

I am trying to understand how copying a pandas data frame works. When I assign a copy of an object in python I am not used to changes to the original object affecting copies of that object. For example:

x = 3
y = x
x = 4
print(y)
3

While x has subsequently been changed, y remains the same. In contrast, when I make changes to a pandas df after assigning it to a copy df1 the copy is also affected by changes to the original DataFrame.

import pandas as pd
import numpy as np

def minusone(x):
    return int(x) - 1

df = pd.DataFrame({"A": [10,20,30,40,50], "B": [20, 30, 10, 40, 50], "C": [32, 234, 23, 23, 42523]})

df1 = df


print(df1['A'])

0    10
1    20
2    30
3    40
4    50
Name: A, dtype: int64

df['A'] = np.vectorize(minusone)(df['A'])

print(df1['A'])

0     9
1    19
2    29
3    39
4    49
Name: A, dtype: int64

The solution appears to be making a deep copy with copy.deepcopy() , but because this behavior is different from the behavior I am used to in python I was wondering if someone could explain what the reasoning behind this difference is or if it is a bug.

Answer 1

In your first example, you did not make a change to the value of x . You assigned a new value to x .

In your second example, you did modify the value of df , by changing one of its columns.

You can see the effect with builtin types too:

>>> x = []
>>> y = x
>>> x.append(1)
>>> y
[1]

The behavior is not specific to Pandas; it is fundamental to Python. There are many, many questions on this site about this same issue, all stemming from the same misunderstanding. The syntax

barename = value

does not have the same behavior as any other construct in Python .

When using name[key] = value , or name.attr = value or name.methodcall() , you may be mutating the value of the object referred to by name , you may be copying something, etc. By using name = value (where name is a single identifier, no dots, no brackets, etc.), you never mutate anything, and never copy anything.

In your first example, you used the syntax x = ... . In your second example, you used the syntax df['A'] = ... . These are not the same syntax, so you can't assume they have the same behavior.

The way to make a copy depends on the kind of object you're trying to copy. For your case, use df1 = df.copy() .