Pandas 重采样不规则时间序列

Question

I have a time series that looks something like this:

2018-10-12 00:00:00 1
2018-10-12 01:00:00 0
2018-10-12 02:00:00 0
2018-10-12 06:00:00 7
2018-10-12 07:00:00 22
2018-10-12 08:00:00 8
2018-10-12 09:00:00 18
2018-10-12 10:00:00 24
2018-10-12 11:00:00 8
2018-10-12 11:15:00 5
2018-10-12 11:30:00 4
2018-10-12 11:45:00 25
2018-10-12 12:00:00 29
2018-10-12 12:15:00 19
2018-10-12 12:30:00 24
2018-10-12 12:45:00 16
2018-10-12 13:00:00 49
2018-10-12 14:00:00 36
2018-10-12 15:00:00 27
2018-10-12 16:00:00 20
2018-10-12 17:00:00 8
2018-10-12 17:15:00 7
2018-10-12 17:30:00 8
2018-10-12 17:45:00 9
2018-10-12 18:00:00 10

I would like to resample it, so that it has 15 minute intervals.

import pandas as pd

data = pd.read_csv("data.csv", sep=",", index_col=0, parse_dates=True)

data_resampled = data.resample("900s").sum()

That yields this result:

2018-10-12 07:00:00 22
2018-10-12 07:15:00 0
2018-10-12 07:30:00 0
2018-10-12 07:45:00 0
2018-10-12 08:00:00 8
2018-10-12 08:15:00 0
2018-10-12 08:30:00 0
2018-10-12 08:45:00 0

But the result I want is:

2018-10-12 07:00:00 5,5
2018-10-12 07:15:00 5,5
2018-10-12 07:30:00 5,5
2018-10-12 07:45:00 5,5
2018-10-12 08:00:00 2
2018-10-12 08:15:00 2
2018-10-12 08:30:00 2
2018-10-12 08:45:00 2

Or ideally something like this

2018-10-12 07:00:00 6
2018-10-12 07:15:00 5
2018-10-12 07:30:00 6
2018-10-12 07:45:00 5
2018-10-12 08:00:00 2
2018-10-12 08:15:00 2
2018-10-12 08:30:00 2
2018-10-12 08:45:00 2

But I will settle for something like this:

2018-10-12 07:00:00 5
2018-10-12 07:15:00 5
2018-10-12 07:30:00 5
2018-10-12 07:45:00 5
2018-10-12 08:00:00 2
2018-10-12 08:15:00 2
2018-10-12 08:30:00 2
2018-10-12 08:45:00 2

How do I resample so that an interval that spans multiple of the new intervals is divided equally, or close to equal across the new smaller intervals?

Answer 1

what you can do is in the resample.sum and use min_count=1 to put the value to NaN if there was no value for this 15min interval before. then you can groupby.transform per group starting where a value exists with notna and cumsum (if a value is followed by nan then they are grouped together), and use mean in the transform with fillna the nan with 0 before.

s_ = s.resample('15min').sum(min_count=1)
s_ = s_.fillna(0).groupby(s_.notna().cumsum()).transform('mean')

print (s_)
2018-10-12 00:00:00     0.25 #here it is 1 divided by 4
2018-10-12 00:15:00     0.25
2018-10-12 00:30:00     0.25
2018-10-12 00:45:00     0.25
2018-10-12 01:00:00     0.00
...
2018-10-12 07:00:00     5.50 #same here
2018-10-12 07:15:00     5.50
2018-10-12 07:30:00     5.50
2018-10-12 07:45:00     5.50
2018-10-12 08:00:00     2.00
...
2018-10-12 17:00:00     8.00 # here you keep the original value as existed before
2018-10-12 17:15:00     7.00 
2018-10-12 17:30:00     8.00
2018-10-12 17:45:00     9.00
2018-10-12 18:00:00    10.00
Freq: 15T, Name: val, dtype: float64

where s would be a series s=data['name_col_to_resample']

Answer 2

I would do resample('H').sum() , then do a asfreq('15Min') followed by groupby :

s = df.resample('H').sum().asfreq('15Min').fillna(0)
s.groupby(s.index.floor('H')).transform('mean')

Output (head):

                        1
0                        
2018-10-12 00:00:00  0.25
2018-10-12 00:15:00  0.25
2018-10-12 00:30:00  0.25
2018-10-12 00:45:00  0.25
2018-10-12 01:00:00  0.00

Answer 3

Try this

import pandas as pd

data = pd.read_csv("data.csv", sep=",", index_col=0, parse_dates=True)
# just changing the column names
df.index.name='Datetime' 
df.columns = ['values']

# resample
df = df.resample('15min').sum().reset_index() # resample

# This will be used for the groupby
df['key'] = np.cumsum( (df['Datetime'].dt.minute == 0) | (df['values'] > 0) )

df['new_values'] = df.groupby(['key'])['values'].transform('mean')

df = df.drop(columns=['key'])

Notice that when you have the following case

2018-10-12 08:00:00 10
2018-10-12 08:15:00 9
2018-10-12 08:30:00 0
2018-10-12 08:45:00 0

it will become

2018-10-12 08:00:00 10
2018-10-12 08:15:00 3
2018-10-12 08:30:00 3
2018-10-12 08:45:00 3

i dont know if this is what you want.