python Pandas DataFrame copy(deep=False) vs copy(deep=True) vs '='

If you see the object IDs of the various DataFrames you create, you can clearly see what is happening.

When you write df2 = df1, you are creating a variable named df2, and binding it with an object with id 4541269200. When you write df1 = pd.DataFrame([9,9,9]), you are creating a new object with id 4541271120 and binding it to variable df1, but the object with id 4541269200 which was previously bound to df1 continues to live. If there were no variables bound to that object, it will get garbage collected by Python.

In[33]: import pandas as pd
In[34]: df1 = pd.DataFrame([1,2,3,4,5])
In[35]: id(df1)
Out[35]: 4541269200

In[36]: df2 = df1
In[37]: id(df2)
Out[37]: 4541269200  # Same id as df1

In[38]: df3 = df1.copy()
In[39]: id(df3)
Out[39]: 4541269584  # New object, new id.

In[40]: df4 = df1.copy(deep=False)
In[41]: id(df4)
Out[41]: 4541269072  # New object, new id.

In[42]: df1 = pd.DataFrame([9, 9, 9])
In[43]: id(df1)
Out[43]: 4541271120  # New object created and bound to name 'df1'.

In[44]: id(df2)
Out[44]: 4541269200  # Old object's id not impacted.

Edit: Added on 7/30/2018

Deep copying doesn’t work in pandas and the devs consider putting mutable objects inside a DataFrame as an antipattern. Consider the following:

In[10]: arr1 = [1, 2, 3]
In[11]: arr2 = [1, 2, 3, 4]
In[12]: df1 = pd.DataFrame([[arr1], [arr2]], columns=['A'])
In[13]: df1.applymap(id)
Out[13]: 
            A
0  4515714832
1  4515734952

In[14]: df2 = df1.copy(deep=True)
In[15]: df2.applymap(id)
Out[15]: 
            A
0  4515714832
1  4515734952

In[16]: df2.loc[0, 'A'].append(55)
In[17]: df2
Out[17]: 
               A
0  [1, 2, 3, 55]
1   [1, 2, 3, 4]
In[18]: df1
Out[18]: 
               A
0  [1, 2, 3, 55]
1   [1, 2, 3, 4]

df2, if it was a true deep copy should have had new ids for the lists contained within it. As a result, when you modify a list inside df2, it affects the list inside df1 as well, because they are the same objects.