Why does date format change with np.where and how do you stop it from happening-CodePudding

I have the following subset of data from a dataframe.

{'NID': {131598: '215026851',
  131599: '215026851',
  131600: '215026851',
  131601: '215026851',
  131602: '215026851',
  131603: '215026851',
  131604: '215026851',
  131605: '215026851',
  131606: '215026851'},
 'AbCode': {131598: 0,
  131599: 0,
  131600: 0,
  131601: 0,
  131602: 0,
  131603: 1,
  131604: 0,
  131605: 0,
  131606: 0},
 'ABdat': {131598: Timestamp('2018-01-24 00:00:00'),
  131599: Timestamp('2019-01-25 00:00:00'),
  131600: NaT,
  131601: Timestamp('2019-11-08 00:00:00'),
  131602: Timestamp('2020-01-24 00:00:00'),
  131603: Timestamp('2020-02-15 00:00:00'),
  131604: Timestamp('2020-10-16 00:00:00'),
  131605: Timestamp('2020-10-26 00:00:00'),
  131606: NaT}}

When formatted the data looks like below

          NID     AbCode  ABdat
131598  215026851   0   2018-01-24
131599  215026851   0   2019-01-25
131600  215026851   0   NaT
131601  215026851   0   2019-11-08
131602  215026851   0   2020-01-24
131603  215026851   1   2020-02-15
131604  215026851   0   2020-10-16
131605  215026851   0   2020-10-26
131606  215026851   0   NaT

I would like to replace the ABdat with missing (NaT) for AbCode = 0 and replace ABdat with ABdat-7days for AbCode = 1

I wrote the following np.where code below to do this.

breed_info['ABdat'] = np.where(breed_info.AbCode == 1, breed_info['ABdat'] - pd.DateOffset(days=7), breed_info['ABdat'].isnull)

The output is presented below

          NID   AbCode       ABdat
131598  215026851   0   <bound method Series.isnull of 49017 ...
131599  215026851   0   <bound method Series.isnull of 49017 ...
131600  215026851   0   <bound method Series.isnull of 49017 ...
131601  215026851   0   <bound method Series.isnull of 49017 ...
131602  215026851   0   <bound method Series.isnull of 49017 ...
131603  215026851   1   1581120000000000000
131604  215026851   0   <bound method Series.isnull of 49017 ...
131605  215026851   0   <bound method Series.isnull of 49017 ...
131606  215026851   0   <bound method Series.isnull of 49017 ...

Could you please advise why the date format is changing and how I can avoid this from happening?

Thanks

CodePudding user response：

Simpluiest is use some pandas solutions with pandas method e.g. Series.where:

breed_info['ABdat'] = (breed_info['ABdat'] - pd.DateOffset(days=7))
                                                        .where(breed_info.AbCode == 1)

With np.where hacky solution with helper Series:

breed_info['ABdat'] = np.where(breed_info.AbCode == 1,
                               breed_info['ABdat'] - pd.DateOffset(days=7),
                                pd.Series(pd.NaT, index=breed_info.index))
print (breed_info)
              NID  AbCode      ABdat
131598  215026851       0        NaT
131599  215026851       0        NaT
131600  215026851       0        NaT
131601  215026851       0        NaT
131602  215026851       0        NaT
131603  215026851       1 2020-02-08
131604  215026851       0        NaT
131605  215026851       0        NaT
131606  215026851       0        NaT

because if passing pd.NAT it return underline numpy array (in nanoseconds):

breed_info['ABdat'] = np.where(breed_info.AbCode == 1, 
                               breed_info['ABdat'] - pd.DateOffset(days=7),
                               pd.NaT)
print (breed_info)
              NID  AbCode                ABdat
131598  215026851       0                  NaT
131599  215026851       0                  NaT
131600  215026851       0                  NaT
131601  215026851       0                  NaT
131602  215026851       0                  NaT
131603  215026851       1  1581120000000000000
131604  215026851       0                  NaT
131605  215026851       0                  NaT
131606  215026851       0                  NaT

I think reason is bug.