convert days difference to numeric values in python pandas

Question

I have a column 'datedif' in my dataframe as :

  exposuredate min_exposure_date    datedif
  2014-10-08   2014-09-27           11 days
  2014-10-09   2014-09-27           12 days
  2014-09-27   2014-09-27           0 days
  2014-09-27   2014-09-27           0 days
  2014-10-22   2014-09-27           25 days

  data.exposuredate = pd.to_datetime(data.exposuredate)
  data.min_exposure_date = pd.to_datetime(data.min_exposure_date)

  data['datedif'] = ((data.exposuredate)-(data.min_exposure_date))

The format for the columns are datetime64[ns]. I want to extract the number of days in the field 'datedif'. I am not able to find anything which could help me to extract the difference in terms of number of days.

I tried:

data['datedif_day'] = data['datedif'].dt.days

Error:

AttributeError: 'Series' object has no attribute 'dt'

Answer 1

The Pandas Docs relate to the types of conversions you are looking for as Frequency Conversion

The two options are 1) division by Timedelta or 2) type conversion. There is a subtle difference between the two as stated in the docs:

"Note that division by the numpy scalar is true division, while astyping is equivalent of floor division."

data = pd.DataFrame([("2014-10-08", "2014-09-27"),
                     ("2014-10-09", "2014-09-27"),
                     ("2014-09-27", "2014-09-27"),
                     ("2014-10-22", "2014-09-27")],
                    columns=["exposuredate", "min_exposure_date"])

data['datediff'] =   pd.to_datetime(data.exposuredate) 
                   - pd.to_datetime(data.min_exposure_date)

data['datediff'] / pd.Timedelta(1, unit='d')
data['datediff'].astype('timedelta64[D]')

Both operations yield:

0    11.0
1    12.0
2     0.0
3    25.0
Name: datediff, dtype: float64

If you are using the date difference as a feature for training a machine learning algorithm, it doesn't matter in which form they are represented as they should be normalised anyway. timedelta64[ns] is a perfectly fine for that. When it comes to visualisation purposes, see this post .

Answer 2

The 'datedif' looks in days format but actually it is in seconds. So in order to get number of days for furhter use add the following line in the code:

   data['datedif'] = data['datedif'].astype(np.numpy64)
   data['datedif_day'] = (data['datedif']/86400000000000)

Answer 3

Came across this same question today and I think the following solutin is the easiest:

Setup:

df = pd.DataFrame([("2014-10-08", "2014-09-27"),
                     ("2014-10-09", "2014-09-27"),
                     ("2014-09-27", "2014-09-27"),
                     ("2014-10-22", "2014-09-27")],
                    columns=["exposuredate", "min_exposure_date"])

df['datediff'] =   pd.to_datetime(df.exposuredate) - pd.to_datetime(df.min_exposure_date)

    exposuredate    min_exposure_date   datediff
0   2014-10-08      2014-09-27          11 days
1   2014-10-09      2014-09-27          12 days
2   2014-09-27      2014-09-27          0 days
3   2014-10-22      2014-09-27          25 days

Solution:

df.datediff.apply(lambda x: x.days)

0    11
1    12
2     0
3    25
Name: datediff, dtype: int64