Using .ix, .iloc, .loc, .at and .iat to access a DataFrame

Using .iloc

.iloc uses integers to read and write data to a DataFrame.

First, let’s create a DataFrame:

df = pd.DataFrame({'one': [1, 2, 3, 4, 5],
                   'two': [6, 7, 8, 9, 10],
              }, index=['a', 'b', 'c', 'd', 'e'])

This DataFrame looks like:

   one  two
a    1    6
b    2    7
c    3    8
d    4    9
e    5   10

Now we can use .iloc to read and write values. Let’s read the first row, first column:

print df.iloc[0, 0]

This will print out:

We can also set values. Lets set the second column, second row to something new:

df.iloc[1, 1] = '21'

And then have a look to see what happened:

print df 

   one two
a    1   6
b    2  21
c    3   8
d    4   9
e    5  10

Using .loc

.loc uses labels to read and write data.

Let’s setup a DataFrame:

df = pd.DataFrame({'one': [1, 2, 3, 4, 5],
                   'two': [6, 7, 8, 9, 10],
                  }, index=['a', 'b', 'c', 'd', 'e'])

Then we can print the DataFrame to have a look at the shape:

print df

This will output

   one  two
a    1    6
b    2    7
c    3    8
d    4    9
e    5   10

We use the column and row labels to access data with .loc. Let’s set row ‘c’, column ‘two’ to the value 33:

df.loc['c', 'two'] = 33

This is what the DataFrame now looks like:

   one  two
a    1    6
b    2    7
c    3   33
d    4    9
e    5   10

Of note, using df['two'].loc['c'] = 33 may not report a warning, and may even work, however, using df.loc['c', 'two'] is guaranteed to work correctly, while the former is not.

We can read slices of data, for example

print df.loc['a':'c']

will print rows a to c. This is inclusive.

   one  two
a    1    6
b    2    7
c    3    8

And finally, we can do both together:

print df.loc['b':'d', 'two']

Will output rows b to c of column ‘two’. Notice that the column label is not printed.

b    7
c    8
d    9

If .loc is supplied with an integer argument that is not a label it reverts to integer indexing of axes (the behaviour of .iloc). This makes mixed label and integer indexing possible:

df.loc['b', 1]

will return the value in 2nd column (index starting at 0) in row ‘b’: