Decode base64 on a python dataframe with missing values

Question

I have a dataframe with a column val_string that is sometimes populated with a base64 encoded string, and other times is NaN.

df
    base_type   val_int     val_string
0     integer        34            NaN
1      string       NaN   c3RyaW5nMQ==
2     integer       108            NaN
3     integer      3586            NaN
4      string       NaN   c3RyaW5nMg==

How do I apply base64.b64decode to only the rows that have a val_string that is not NaN?

I tried this, only to get a strange OSError: could not get source code:

df['val_string'] = df['val_string'].apply(lambda x: df['val_string'] if pd.isna(df['val_string']) else base64.b64decode(x))

Any help would be much appreciated!

Answer 1

Use boolean indexing:

from base64 import b64decode

m = df['val_string'].notna()
df.loc[m, 'val_string'] = df.loc[m, 'val_string'].apply(b64decode)

Or with your approach:

from base64 import b64decode

df['val_string'] = df['val_string'].apply(lambda x: x if pd.isna(x)
                                          else b64decode(x))

Output:

  base_type  val_int  val_string
0   integer     34.0         NaN
1    string      NaN  b'string1'
2   integer    108.0         NaN
3   integer   3586.0         NaN
4    string      NaN  b'string2'