How to remove levels from a multi-indexed dataframe?

Question

For example, I have:

In [1]: df = pd.DataFrame([8, 9],
                          index=pd.MultiIndex.from_tuples([(1, 1, 1),
                                                           (1, 3, 2)]),
                          columns=['A'])

In [2] df
Out[2]: 
       A
1 1 1  8
  3 2  9

Is there a better way to remove the last level from the index than this:

In [3]: pd.DataFrame(df.values,
                     index=df.index.droplevel(2),
                     columns=df.columns)
Out[3]: 
     A
1 1  8
  3  9

score 105 · Accepted Answer · answered Jun 13 '13 at 10:43

105

df.reset_index(level=2, drop=True)
Out[29]: 
     A
1 1  8
  3  9

answered Jun 13 '13 at 10:43

Yariv

12,945
19
54
75

11

Worth noting, that there is an inplace argument, which is `True` would modify `df` rather than returning a new DataFrame. – Andy Hayden Jun 13 '13 at 11:11
This works right now, even in a `pd.Series`. The `.index.drop(...)` ideas may be obsolete. – Tomasz Gandor Apr 16 '19 at 00:41
If your index level 1 is miningful and not just the perfect range of numbers you will be reseting the level 1 too? – aerijman Jul 01 '21 at 17:26

score 75 · Answer 2 · edited Jan 28 '19 at 21:34

75

You don't need to create a new DataFrame instance! You can modify the index:

df.index = df.index.droplevel(2)
df

     A
1 1  8
  3  9

You can also specify negative indices, for selection from the end:

df.index = df.index.droplevel(-1)

edited Jan 28 '19 at 21:34

cs95

379,657
97
704
746

answered Jun 13 '13 at 10:41

Andy Hayden

359,921
101
625
535

can this be applied to columns for pandas v 0.23+? – Yuca Apr 26 '19 at 12:28
I've just tried it with v 0.22 and worked for columns. I like this answer better than the accepted one. It is clearer for me, and can be applied to columns as well. – Arpad Horvath -- Слава Україні Jul 10 '19 at 07:54

score 12 · Answer 3 · edited Feb 19 '23 at 23:51

12

If your index has names like

Then you can also remove by specifying the index name

df.index = df.index.droplevel('Z')

edited Feb 19 '23 at 23:51

wjandrea

28,235
9
60
81

answered Feb 23 '19 at 13:12

shashank2806

480
6
14

score 2 · Answer 4 · edited Feb 19 '23 at 23:55

From 0.24+, we can directly droplevel on df. So, to drop the last level of the index:

>>> df

         col
1 5 1 4  foo
  3 2 8  bar
2 4 3 7  saz

>>> df.droplevel(-1)

       col
1 5 1  foo
  3 2  bar
2 4 3  saz

The axis whose levels are dropped can also be controlled with axis argument and it defaults to 0, i.e., over index. Multiple levels can be dropped at once via supplying a list and if any of the index has a name, those can be used, too (as exemplified in the linked doc).

Note: the argument to droplevel is tried to be first interpreted as a label; so if any of the levels happens to have an integer name, it will be dropped i.e., not positionally:

>>> df
                 col
this -1 other 0
1    5  1     4  foo
     3  2     8  bar
2    4  3     7  saz


# literally drops `-1` level
>>> df.droplevel(-1)

              col
this other 0
1    1     4  foo
     2     8  bar
2    3     7  saz

# literally level `0` is dropped
>>> df.droplevel(0)

               col
this -1 other
1    5  1      foo
     3  2      bar
2    4  3      saz

To make sure a positional dropping happens, we can go for the names attribute and select positionally there:

>>> df
                 col
this -1 other 0
1    5  1     4  foo
     3  2     8  bar
2    4  3     7  saz

# go get the name of the last level, drop whatever it is
>>> df.droplevel(df.index.names[-1])

               col
this -1 other
1    5  1      foo
     3  2      bar
2    4  3      saz

# similarly...
>>> df.droplevel(df.index.names[0])

            col
-1 other 0
5  1     4  foo
3  2     8  bar
4  3     7  saz

Lastly, droplevel returns a new dataframe, so df = df.droplevel(...) is needed to see the change in df.

How to remove levels from a multi-indexed dataframe?

4 Answers4

Linked

Related