Pandas 分配意外行为

Question

I am just playing with pandas, trying to modify values of a column.

My initial dataframe is:

df = pd.DataFrame(
    dict(x=[1, 2, 3, 4, 5, 6, 7], y=[10, 11, 15, 14, 14, 25, 25)
    )
df.index = list('abcdefg')

with output:

>>> df
       x   y
    a  1  10
    b  2  11
    c  3  15
    d  4  14
    e  5  14
    f  6  25
    g  7  25

Suppose that I want to modify the first element of x column. I do:

df.loc['a', 'x'] = 100

which outputs:

>>> df.loc['a', 'x'] = 100
>>> df
     x   y
a  100  10
b    2  11
c    3  15
d    4  14
e    5  14
f    6  25
g    7  25

What I can't understand is why the following:

>>> j = df['x']
>>> j['a'] = 200
>>> df
     x   y
a  200  10
b    2  11
c    3  15
d    4  14
e    5  14
f    6  25
g    7  25

also modifies the first element of x column in df . Furthermore:

>>> df.loc['a', 'x'] is j['a']
False

which means that they don't point to the same object. What is going on?

Answer 1

You are not performing the correct test. You should rather test:

j is df['x']

output: True

j and df['x'] point to the same Series.

The False is explained by the underlying numpy array that does not contain python objects. The object are generated during slicing:

import numpy as np

a = np.array([1, 2, 3])
a[0] is a[0]

output: False

Answer 2

That is why we need copy here

j = df['x'].copy()

Notice after add copy the id number is different

id(df['x'])
Out[612]: 140536670316496
id(df['x'].copy())
Out[613]: 140536673228496

Answer 3

Because j = df['x'] only assigns the object reference of df['x'] to j . Anything modified through j will impact the same object in memory as the one behind df['x'] .

If you want to make a copy of the object value behind df['x'] , store it into j , and modify it independently from df , you need to use .copy() method as:

j = df['x'].copy()

Read more about it at https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.copy.html