Convert column values to lower case only if they are string

Question

I'm having real trouble converting a column into lowercase. It's not as simple as just using:

df['my_col'] = df['my_col'].str.lower()

because I'm iterating over a lot of dataframes, and some of them (but not all) have both strings and integers in the column of interest. This causes the lower function, if applied like above, to throw an exception:

AttributeError: Can only use .str accessor with string values, which use np.object_ dtype in pandas

Rather than forcing the type to be a string, I'd like to assess whether values are strings and then - if they are - convert them to lowercase, and - if they are not strings - leave them as they are. I thought this would work:

df = df.apply(lambda x: x.lower() if(isinstance(x, str)) else x)

But it doesn't work... probably because I'm overlooking something obvious, but I can't see what it is!

My data looks something like this:

                          OS    Count
0          Microsoft Windows     3
1                   Mac OS X     4
2                      Linux     234
3    Don't have a preference     0
4  I prefer Windows and Unix     3
5                       Unix     2
6                        VMS     1
7         DOS or ZX Spectrum     2

Answer 1

The test in your lambda function isn't quite right, you weren't far from the truth though:

df.apply(lambda x: x.str.lower() if(x.dtype == 'object') else x)

With the data frame and output:

>>> df = pd.DataFrame(
    [
        {'OS': 'Microsoft Windows', 'Count': 3},
        {'OS': 'Mac OS X', 'Count': 4},
        {'OS': 'Linux', 'Count': 234},
        {'OS': 'Dont have a preference', 'Count': 0},
        {'OS': 'I prefer Windows and Unix', 'Count': 3},
        {'OS': 'Unix', 'Count': 2},
        {'OS': 'VMS', 'Count': 1},
        {'OS': 'DOS or ZX Spectrum', 'Count': 2},
    ]
)
>>> df = df.apply(lambda x: x.str.lower() if x.dtype=='object' else x)
>>> print(df)
                          OS  Count
0          microsoft windows      3
1                   mac os x      4
2                      linux    234
3     dont have a preference      0
4  i prefer windows and unix      3
5                       unix      2
6                        vms      1
7         dos or zx spectrum      2

Answer 2

This also works and is very readable:

for column in df.select_dtypes("object").columns:
    df[column] = df[column].str.lower()

A possible drawback might be the for loop over a subset of columns.