在python pandas中将不规则时间序列转换为每小时数据

Question

我有一个如下所示的数据框：

                read            value
0     2013-01-07 05:00:00        29.0
1     2013-01-08 15:00:00      4034.0
2     2013-01-09 20:00:00    256340.0
3     2013-01-10 20:00:00    343443.0
4     2013-01-11 20:00:00    4642435.0
5     2013-01-12 15:00:00    544296.0
6     2013-01-13 20:00:00    700000.0
7     2013-01-14 20:00:00    782335.0
8     2013-01-15 19:00:00    900000.0
9     2013-01-16 20:00:00    959130.0
10    2013-01-17 19:00:00   1114343.0
11    2013-01-18 20:00:00   1146230.0
12    2013-01-19 20:00:00   1247793.0
13    2013-01-20 20:00:00   1343376.0

我想对其进行转换和规范化，以便它显示一段时间内的每小时消耗量。到目前为止，我有以下内容

import numpy as np
import pandas as pd

#caluclates hourly delta
current['hour_delta'] = (current['read'] - current['read'].shift()).fillna(0).astype('timedelta64[h]')


#adds end date and then amount per hours
current['end_date'] = current['read'] + pd.to_timedelta(current['hour_delta'], unit='h')
current['infer_hour'] = current['value'] / current['hour_delta']

然后我创建系列

#create hourly time series
result = pd.Series(0, index=pd.date_range(start=current['read'].min(), end=current['read'].max(), freq='h'))

但是，从这里我无法弄清楚如何将小时费率应用于该系列。

Answer 1

您可以在 read 列上每小时重新采样一次。然后进行插值以填充空值。然后将每一行与下一行的差异。

例如，2013-01-07 05:00:00 和 2013-01-08 15:00:00 之间有 34 小时。如果我必须在 34 小时内分发 4034，那么每个小时的平均值应该是 4034 / 34 或 118.647059

current.set_index('read').value.cumsum().resample('H').sum().interpolate().diff()

read
2013-01-07 05:00:00             NaN
2013-01-07 06:00:00      118.647059
2013-01-07 07:00:00      118.647059
2013-01-07 08:00:00      118.647059
2013-01-07 09:00:00      118.647059
2013-01-07 10:00:00      118.647059
2013-01-07 11:00:00      118.647059
2013-01-07 12:00:00      118.647059
2013-01-07 13:00:00      118.647059
2013-01-07 14:00:00      118.647059
2013-01-07 15:00:00      118.647059
2013-01-07 16:00:00      118.647059
2013-01-07 17:00:00      118.647059
2013-01-07 18:00:00      118.647059
2013-01-07 19:00:00      118.647059
...