Convert non-numbers in dataframe to NaN (numpy)?

Question

How to convert non-numbers in DataFrame to NaN (numpy)? For example, here is a DataFrame:

a    b
--------
10   ...
4    5
...  6

How to covert it to:

a    b
--------
10   NaN
4    5
NaN  6

Should be marked as duplicate to http://stackoverflow.com/questions/18434208/pandas-converting-to-numeric-creating-nans-when-necessary — pansen, Mar 27 '17 at 14:57

EdChum · Accepted Answer · 2017-03-27T14:58:52.747

10

IIUC you can just do

df = df.apply(lambda x: pd.to_numeric(x, errors='coerce') )

This will force the duff values to NaN, note that the presence of NaN will change the dtype to float as NaN cannot be represented by int

In [6]:
df = df.apply(pd.to_numeric, errors='coerce')
df

Out[6]:
      a    b
0  10.0  NaN
1   4.0  5.0
2   NaN  6.0

The lambda isn't necessary but it's more readable IMO

edited Mar 27 '17 at 14:58

answered Mar 27 '17 at 14:56

EdChum

I like `df = df.apply(pd.to_numeric, errors='coerce')` `apply` will pass the `errors` parameter. – piRSquared Mar 27 '17 at 14:57
is lambda necessary? – AsheKetchum Mar 27 '17 at 14:57
@piRSquared sure, have posted, It's a habit of mine – EdChum Mar 27 '17 at 14:59
1

@AsheKetchum no, it's not necessary but it's a habit of mine for readability – EdChum Mar 27 '17 at 15:00
@EdChum That is true :) – AsheKetchum Mar 27 '17 at 15:01
@AsheKetchum the `lambda` is just syntax. You would need it, if `apply` didn't pass arguments to the function being passed. And I agree with EdChum... the lambda makes it more obvious what is happening. – piRSquared Mar 27 '17 at 15:02

score 4 · Answer 2 · answered Mar 27 '17 at 14:59

4

You can can also stack then unstack the dataframe

pd.to_numeric(df.stack(), errors='coerce').unstack()

      a    b
0  10.0  NaN
1   4.0  5.0
2   NaN  6.0

answered Mar 27 '17 at 14:59

piRSquared

2 Answers2