How to set unexpected data type to na?

Question

I have a data frame like this:

a = pd.DataFrame({'foo':[1,2,3,'str']})

    foo
0   1
1   2
2   3
3   str

I want to set the data type to int64:

a['foo'].astype('int32')

but I got an error message:

ValueError: invalid literal for int() with base 10: 'str'

How to set unexpected data type to NA. In my case, I'd like to return data frame like the following:

Possible duplicate of [How to replace all non-numeric entries with NaN in a pandas dataframe?](https://stackoverflow.com/questions/41938549/how-to-replace-all-non-numeric-entries-with-nan-in-a-pandas-dataframe) — Georgy, Mar 13 '18 at 12:33
You cannot have `dtype 'int32'` and `NA` values at the same time in a `Series`. — Stop harming Monica, Mar 13 '18 at 12:43

jezrael · Accepted Answer · 2018-03-13T12:35:16.157

The best is convert all values to floats, because NaNs are float by to_numeric with parameter errors='coerce':

df = pd.to_numeric(df['foo'], errors='coerce')
print (df)
0    1.0
1    2.0
2    3.0
3    NaN
Name: foo, dtype: float64

But if really need integers with floats, is possible this hack:

df = df['foo'].where(df['foo'].apply(lambda x: isinstance(x, int)))
print (df)
0      1
1      2
2      3
3    NaN
Name: foo, dtype: object

print (df.apply(type))
0      <class 'int'>
1      <class 'int'>
2      <class 'int'>
3    <class 'float'>
Name: foo, dtype: object

score 0 · Answer 2 · answered Mar 13 '18 at 12:56

0

Or using isalpha

a.foo.mask(a.foo.str.isalpha().notnull())
Out[331]: 
0      1
1      2
2      3
3    NaN
Name: foo, dtype: object

answered Mar 13 '18 at 12:56

BENY

2 Answers2