pandas pd.DataFrame.diff(axis=1) NotImplementationError

Question

I have a dataframe that generates from pd.pivot_table with columns having datetime.

I want to calculate the time difference along columns. So I used df.diff(axis=1). I got NotImplementedError.

I guess the problem has been solved? Why I still get NonImplementedError?

Data Sample: Please ignore the "Text". Typo.

What version of pandas are you running? Maybe just need to upgrade? — Chris Adams, Mar 09 '20 at 07:57
Hmmm, I'm using the same. Working for me with datetime types. What about versions for `numpy` and `python-dateutil` ? — Chris Adams, Mar 09 '20 at 08:03
I also have some NaT value in the columns. Would that affect? — JOHN, Mar 09 '20 at 08:06
I'm not able to reproduce with `NaT` - appears to work ok for me — Chris Adams, Mar 09 '20 at 08:12
@ChrisA I got it works by removing the timezone: `data['date'] = data['date'].dt.tz_convert(None)` — JOHN, Mar 09 '20 at 08:22
Good find! You should add as an answer, could be useful for others. — Chris Adams, Mar 09 '20 at 08:23
I also got 88424 days when the previous column is NaT. Any hints? — JOHN, Mar 09 '20 at 08:43
In the context of your data, does it make sense to forward fill to deal with `NaT` ? : `df.ffill(axis=1)` — Chris Adams, Mar 09 '20 at 08:45

score 2 · Accepted Answer · answered Mar 09 '20 at 08:31

2

I tried to remove the timezone in datetime64[ns, UTC] columns. It works!

data['date'] = data['date'].dt.tz_convert(None)

answered Mar 09 '20 at 08:31

JOHN

gosh, i appreciate the ability to work with timezones, but they really screw up all the formatting. i wish the default would have stayed at not having them. suddenly, CSV readings produce different outputs... :( – K.-Michael Aye Mar 27 '20 at 02:48

1 Answers1