If object A is a copy of object B, then each object will be allocated their own memory block for its data. This means that modifying the copy will not mutate the original data and vice versa.

If object A is a view of object B, then they both share a single memory block for their data. This means that modifying the copy will mutate the original data and vice versa.

In the context of Pandas, when you access values of Series or a DataFrame, what is returned can either be a copy or view. This distinction is important for two reasons:

if you don't know whether the return value is a copy or view, then you will not know what happens when you modify the return value - will it mutate the original data or not?
if you're dealing with large datasets, then you might not want the return value to be a copy since copies take up more memory.

Accessing values of a Series/DataFrame

Unfortunately, when you access values from a Series or a DataFrame, the rule that decides whether a copy or view is returned is quite complicated.

Here is a general rule of thumb:

if you access a single column, then a view is returned (e.g. df["A"]).
if you access multiple columns, then a copy is returned (e.g. df[["A","B"]])

Here's a quick demo - consider the following DataFrame:


        
        
            
                
                
                    df = pd.DataFrame({"A":[3,4],"B":[5,6]})
df
                
            
               A  B
0  3  5
1  4  6

Getting a view

To illustrate that accessing a single column returns a view:


        
        
            
                
                
                    col_A = df["A"]   # col_A is a view
col_A[0] = 9
df
                
            
               A  B
0  9  6
1  4  7

Notice how modifying col_A mutated the original df. Note that modifying df will also mutate col_A.

Getting a copy

To illustrate that accessing multiple columns returns a copy:


        
        
            
                
                
                    cols_A_B = df[["A","B"]]   # cols_A_B is a copy
cols_A_B.iloc[0,0] = 9
df
                
            
               A  B
a  3  6
b  4  7

Notice how df did not get mutated.

Other cases

For other cases, whether a copy or view is returned depends on the situation. Whenever in doubt, it is good practise to use the _is_view property to verify:


        
        
            
                
                
                    df = pd.DataFrame({"A":[3,4],"B":[6,7]})
df[["A","B"]]._is_view
                
            
            False

Methods that can return copies

Many Pandas methods allow you to specify whether you want a view or a copy. For instance, consider the method to_numpy(~), which returns a NumPy array representation of a DataFrame.

By default, copy=False, which means that modifying the returned NumPy array would mutate the original DataFrame:


        
        
            
                
                
                    numpy_view = df.to_numpy()   # copy=False
numpy_view[0,0] = 9          # modify top-left entry
df
                
            
               A  B
a  9  6
b  4  7

Note that modifying df would also affect numpy_view.

To get a new copy of NumPy array that is independent of the original DataFrame, set copy=True like so:


        
        
            
                
                
                    numpy_copy = df.to_numpy(copy=True)
numpy_copy[0,0] = 9
df
                
            
               A  B
a  3  6
b  4  7

Notice how modifying the returned value does not mutate the original df.

Copying Pandas object

Pandas has the method copy(~) that makes a copy of a Pandas object:


        
        
            
                
                
                    df = pd.DataFrame({"A":[3,4],"B":[6,7]}, index=["a","b"])
df_copy = df.copy()
df_copy.iloc[0,0] = 10
df
                
            
               A  B
a  3  6
b  4  7

Published by Isshin Inada

Edited by 0 others

Did you find this page useful?

thumb_up

thumb_down

Comment

Citation

Ask a question or leave a feedback...

thumb_up

thumb_down

chat_bubble_outline

settings

Enjoy our search

Hit / to insta-search docs and recipes!