How are iloc and loc different?

AZhao picture AZhao · Jul 23, 2015 · Viewed 517.8k times · Source

Can someone explain how these two methods of slicing are different?
I've seen the docs, and I've seen these answers, but I still find myself unable to explain how the three are different. To me, they seem interchangeable in large part, because they are at the lower levels of slicing.

For example, say we want to get the first five rows of a DataFrame. How is it that these two work?

df.loc[:5]
df.iloc[:5]

Can someone present three cases where the distinction in uses are clearer?


Once upon a time, I also wanted to know how these two functions differ from df.ix[:5] but ix has been removed from pandas 1.0, so I don't care anymore!

Answer

Alex Riley picture Alex Riley · Jul 23, 2015

Note: in pandas version 0.20.0 and above, ix is deprecated and the use of loc and iloc is encouraged instead. I have left the parts of this answer that describe ix intact as a reference for users of earlier versions of pandas. Examples have been added below showing alternatives to ix.


First, here's a recap of the three methods:

  • loc gets rows (or columns) with particular labels from the index.
  • iloc gets rows (or columns) at particular positions in the index (so it only takes integers).
  • ix usually tries to behave like loc but falls back to behaving like iloc if a label is not present in the index.

It's important to note some subtleties that can make ix slightly tricky to use:

  • if the index is of integer type, ix will only use label-based indexing and not fall back to position-based indexing. If the label is not in the index, an error is raised.

  • if the index does not contain only integers, then given an integer, ix will immediately use position-based indexing rather than label-based indexing. If however ix is given another type (e.g. a string), it can use label-based indexing.


To illustrate the differences between the three methods, consider the following Series:

>>> s = pd.Series(np.nan, index=[49,48,47,46,45, 1, 2, 3, 4, 5])
>>> s
49   NaN
48   NaN
47   NaN
46   NaN
45   NaN
1    NaN
2    NaN
3    NaN
4    NaN
5    NaN

We'll look at slicing with the integer value 3.

In this case, s.iloc[:3] returns us the first 3 rows (since it treats 3 as a position) and s.loc[:3] returns us the first 8 rows (since it treats 3 as a label):

>>> s.iloc[:3] # slice the first three rows
49   NaN
48   NaN
47   NaN

>>> s.loc[:3] # slice up to and including label 3
49   NaN
48   NaN
47   NaN
46   NaN
45   NaN
1    NaN
2    NaN
3    NaN

>>> s.ix[:3] # the integer is in the index so s.ix[:3] works like loc
49   NaN
48   NaN
47   NaN
46   NaN
45   NaN
1    NaN
2    NaN
3    NaN

Notice s.ix[:3] returns the same Series as s.loc[:3] since it looks for the label first rather than working on the position (and the index for s is of integer type).

What if we try with an integer label that isn't in the index (say 6)?

Here s.iloc[:6] returns the first 6 rows of the Series as expected. However, s.loc[:6] raises a KeyError since 6 is not in the index.

>>> s.iloc[:6]
49   NaN
48   NaN
47   NaN
46   NaN
45   NaN
1    NaN

>>> s.loc[:6]
KeyError: 6

>>> s.ix[:6]
KeyError: 6

As per the subtleties noted above, s.ix[:6] now raises a KeyError because it tries to work like loc but can't find a 6 in the index. Because our index is of integer type ix doesn't fall back to behaving like iloc.

If, however, our index was of mixed type, given an integer ix would behave like iloc immediately instead of raising a KeyError:

>>> s2 = pd.Series(np.nan, index=['a','b','c','d','e', 1, 2, 3, 4, 5])
>>> s2.index.is_mixed() # index is mix of different types
True
>>> s2.ix[:6] # now behaves like iloc given integer
a   NaN
b   NaN
c   NaN
d   NaN
e   NaN
1   NaN

Keep in mind that ix can still accept non-integers and behave like loc:

>>> s2.ix[:'c'] # behaves like loc given non-integer
a   NaN
b   NaN
c   NaN

As general advice, if you're only indexing using labels, or only indexing using integer positions, stick with loc or iloc to avoid unexpected results - try not use ix.


Combining position-based and label-based indexing

Sometimes given a DataFrame, you will want to mix label and positional indexing methods for the rows and columns.

For example, consider the following DataFrame. How best to slice the rows up to and including 'c' and take the first four columns?

>>> df = pd.DataFrame(np.nan, 
                      index=list('abcde'),
                      columns=['x','y','z', 8, 9])
>>> df
    x   y   z   8   9
a NaN NaN NaN NaN NaN
b NaN NaN NaN NaN NaN
c NaN NaN NaN NaN NaN
d NaN NaN NaN NaN NaN
e NaN NaN NaN NaN NaN

In earlier versions of pandas (before 0.20.0) ix lets you do this quite neatly - we can slice the rows by label and the columns by position (note that for the columns, ix will default to position-based slicing since 4 is not a column name):

>>> df.ix[:'c', :4]
    x   y   z   8
a NaN NaN NaN NaN
b NaN NaN NaN NaN
c NaN NaN NaN NaN

In later versions of pandas, we can achieve this result using iloc and the help of another method:

>>> df.iloc[:df.index.get_loc('c') + 1, :4]
    x   y   z   8
a NaN NaN NaN NaN
b NaN NaN NaN NaN
c NaN NaN NaN NaN

get_loc() is an index method meaning "get the position of the label in this index". Note that since slicing with iloc is exclusive of its endpoint, we must add 1 to this value if we want row 'c' as well.

There are further examples in pandas' documentation here.