How to resample daily data to hourly data for all whole days with pandas?

user_11680166 0
Viewed 1228

I have a dataframe df like the one below:

    city    datetime    value
0   city_a  2020-07-10  2
1   city_a  2020-07-11  5
2   city_b  2020-07-11  4

And I am trying to resample the daily datetimes with a 6h frequency (data every 00h, 6h, 12h and 18h).

The following code gives me almost the output I am expecting

my_df = my_df.set_index(['datetime', 'city'])
my_df = my_df.unstack(-1).resample('6H').pad()
my_df = my_df.stack().reset_index()
my_df = my_df[['city', 'datetime', 'value']]
my_df = my_df.sort_values(['city', 'datetime'])

Output:

    city    datetime            value
0   city_a  2020-07-10 00:00:00 2.0
1   city_a  2020-07-10 06:00:00 2.0
2   city_a  2020-07-10 12:00:00 2.0
3   city_a  2020-07-10 18:00:00 2.0
4   city_a  2020-07-11 00:00:00 5.0
5   city_b  2020-07-11 00:00:00 4.0

However, we can see that the day 2020-07-11 is not complete. I would like the rows including 2020-07-11 06:00:00, 12:00:00 and 18:00:00 to appear into the output.

So my expected output should be:

    city    datetime            value
0   city_a  2020-07-10 00:00:00 2.0
1   city_a  2020-07-10 06:00:00 2.0
2   city_a  2020-07-10 12:00:00 2.0
3   city_a  2020-07-10 18:00:00 2.0
4   city_a  2020-07-11 00:00:00 5.0
6   city_a  2020-07-11 06:00:00 5.0
8   city_a  2020-07-11 12:00:00 5.0
10  city_a  2020-07-11 18:00:00 5.0
5   city_b  2020-07-11 00:00:00 4.0
7   city_b  2020-07-11 06:00:00 4.0
9   city_b  2020-07-11 12:00:00 4.0
11  city_b  2020-07-11 18:00:00 4.0

Is there an elegant way to do it with Pandas ?

Code to generate the dataframe:

my_df = pd.DataFrame(data = {
    'city': ['city_a', 'city_a', 'city_b'],
    'datetime': 
[pd.to_datetime('2020/07/10'),pd.to_datetime('2020/07/11'),pd.to_datetime('2020/07/11')],
    'value': [2,5,4]
})
pandas datetime python
2 Answers

Use:

# STEP A
df1 = (df.groupby('city')['datetime'].max() + pd.Timedelta(days=1)).reset_index()

# STEP B
df1 = pd.concat([df, df1]).set_index('datetime')

# STEP C
df1 = df1.groupby('city', as_index=False).resample('6H').ffill()

# STEP D
df1 = df1.reset_index().drop('level_0', 1).dropna(subset=['value'])

Details:

STEP A: Use DataFrame.groupby to group the dataframe on city to determine the maximum value of date in each group and add 1 day to max value of every group, this will be needed for resampling the dataframe.

# print(df1)
     city   datetime
0  city_a 2020-07-12
1  city_b 2020-07-12

STEP B: Using pd.concat to concat the original dataframe df to the newly created dataframe df1, this is because we have to resample the dataframe in STEP C.

# print(df1)
              city  value
datetime                 
2020-07-10  city_a    2.0
2020-07-11  city_a    5.0
2020-07-11  city_b    4.0
2020-07-12  city_a    NaN
2020-07-12  city_b    NaN

STEP C: Using DataFrame.resample resample the dataframe grouped on city with a frequency of 6H and use ffill to forward fill the values.

# print(df1)
                         city  value
  datetime                          
0 2020-07-10 00:00:00  city_a    2.0
  2020-07-10 06:00:00  city_a    2.0
  2020-07-10 12:00:00  city_a    2.0
  2020-07-10 18:00:00  city_a    2.0
  2020-07-11 00:00:00  city_a    5.0
  2020-07-11 06:00:00  city_a    5.0
  2020-07-11 12:00:00  city_a    5.0
  2020-07-11 18:00:00  city_a    5.0
  2020-07-12 00:00:00  city_a    NaN
1 2020-07-11 00:00:00  city_b    4.0
  2020-07-11 06:00:00  city_b    4.0
  2020-07-11 12:00:00  city_b    4.0
  2020-07-11 18:00:00  city_b    4.0
  2020-07-12 00:00:00  city_b    NaN

STEP D: Finally use DataFrame.reset_index and drop the unused columns using DataFrame.drop along axis=1, also use DataFrame.dropna to drop the rows with NaN values in column value.

# print(df1)
              datetime    city  value
0  2020-07-10 00:00:00  city_a    2.0
1  2020-07-10 06:00:00  city_a    2.0
2  2020-07-10 12:00:00  city_a    2.0
3  2020-07-10 18:00:00  city_a    2.0
4  2020-07-11 00:00:00  city_a    5.0
5  2020-07-11 06:00:00  city_a    5.0
6  2020-07-11 12:00:00  city_a    5.0
7  2020-07-11 18:00:00  city_a    5.0
9  2020-07-11 00:00:00  city_b    4.0
10 2020-07-11 06:00:00  city_b    4.0
11 2020-07-11 12:00:00  city_b    4.0
12 2020-07-11 18:00:00  city_b    4.0
user_128331660

The only way I see is to add a empty row, having a datetime equal to the latest existing datetime + one day. Then you can nearly do exactly the same (pivot is a handy way to replace set_index and unstack).

# adding a row where datetime corresponds to the max datetime + 1 day
df.loc[len(df), 'datetime'] = df.datetime.max() + pd.Timedelta(days=1)

# pivot to replace set_index & unstack
df = (df.pivot(index='datetime', columns='city')
   .resample('6H')
   .pad(3)
   .stack()
   .reset_index()
   .sort_values(['city', 'datetime']))

df[['city', 'datetime', 'value']]

    city    datetime    value
0   city_a  2020-07-10 00:00:00 2.0
1   city_a  2020-07-10 06:00:00 2.0
2   city_a  2020-07-10 12:00:00 2.0
3   city_a  2020-07-10 18:00:00 2.0
4   city_a  2020-07-11 00:00:00 5.0
6   city_a  2020-07-11 06:00:00 5.0
8   city_a  2020-07-11 12:00:00 5.0
10  city_a  2020-07-11 18:00:00 5.0
5   city_b  2020-07-11 00:00:00 4.0
7   city_b  2020-07-11 06:00:00 4.0
9   city_b  2020-07-11 12:00:00 4.0
11  city_b  2020-07-11 18:00:00 4.0
user_137650850
Related