python - 上采样时间，同时保留 python 中的原始点值

Question

我有一个任意大小的不均匀间隔的时间序列(例如 7)，见下文，我想对其进行上采样(50)并保留原始点。

ls = ['2016-01-30 12:10:00', 
  '2016-01-30 12:23:35', 
  '2016-01-30 12:24:14', 
  '2016-01-30 12:24:51', 
  '2016-01-30 12:25:00', 
  '2016-01-30 12:26:49', 
  '2016-01-30 12:27:36']

新时间戳存储在哪个点之间基本上并不重要，但最好根据时间间隔存储。因此时间间隔越大，新生成的时间戳就越多，例如

new_ls = ['2016-01-30 12:10:00',
      x,
      x,
      x,
      x,
  '2016-01-30 12:23:35',
      x,
      x,
  '2016-01-30 12:24:14',
  '2016-01-30 12:24:51',
  '2016-01-30 12:25:00',
      x,
  '2016-01-30 12:26:49',
  '2016-01-30 12:27:36']

结果列表的间距也可能不均匀。提前致谢

Answer 1

首先将列表转换为 pd.DatetimeIndex 并找到开始和结束之间时间戳的理想分布:

import pandas as pd
import numpy as np

ls = pd.to_datetime(['2016-01-30 12:10:00', 
                     '2016-01-30 12:23:35', 
                     '2016-01-30 12:24:14', 
                     '2016-01-30 12:24:51', 
                     '2016-01-30 12:25:00', 
                     '2016-01-30 12:26:49', 
                     '2016-01-30 12:27:36'])

n = 50
dt = (ls.max() - ls.min())/n
ls_temp = pd.date_range(start=ls.min(), end=ls.max(), freq=dt)

然后找出原始测量值的放置位置，例如用最小绝对差替换条目:

idx = np.abs(ls[:,None] - ls_temp[None, :]).argmin(axis=1)
ls_temp = pd.Series(ls_temp)
ls_temp[idx] = pd.Series(ls)

请注意，此策略在某些边缘情况下可能会失败(其中一个索引被两个值或类似情况替换)。

最后，您可以将其转换回您的格式，例如:

ls = list(map("{:%Y-%m-%d %H:%M:%S}".format, ls_temp.tolist()))

输出:

['2016-01-30 12:10:00',
 '2016-01-30 12:10:21',
 '2016-01-30 12:10:42',
 '2016-01-30 12:11:03',
 '2016-01-30 12:11:24',
 '2016-01-30 12:11:45',
 '2016-01-30 12:12:06',
 '2016-01-30 12:12:27',
 '2016-01-30 12:12:48',
 '2016-01-30 12:13:10',
 '2016-01-30 12:13:31',
 '2016-01-30 12:13:52',
 '2016-01-30 12:14:13',
 '2016-01-30 12:14:34',
 '2016-01-30 12:14:55',
 '2016-01-30 12:15:16',
 '2016-01-30 12:15:37',
 '2016-01-30 12:15:59',
 '2016-01-30 12:16:20',
 '2016-01-30 12:16:41',
 '2016-01-30 12:17:02',
 '2016-01-30 12:17:23',
 '2016-01-30 12:17:44',
 '2016-01-30 12:18:05',
 '2016-01-30 12:18:26',
 '2016-01-30 12:18:48',
 '2016-01-30 12:19:09',
 '2016-01-30 12:19:30',
 '2016-01-30 12:19:51',
 '2016-01-30 12:20:12',
 '2016-01-30 12:20:33',
 '2016-01-30 12:20:54',
 '2016-01-30 12:21:15',
 '2016-01-30 12:21:36',
 '2016-01-30 12:21:58',
 '2016-01-30 12:22:19',
 '2016-01-30 12:22:40',
 '2016-01-30 12:23:01',
 '2016-01-30 12:23:22',
 '2016-01-30 12:23:35',
 '2016-01-30 12:24:14',
 '2016-01-30 12:24:25',
 '2016-01-30 12:24:51',
 '2016-01-30 12:25:00',
 '2016-01-30 12:25:29',
 '2016-01-30 12:25:50',
 '2016-01-30 12:26:11',
 '2016-01-30 12:26:32',
 '2016-01-30 12:26:49',
 '2016-01-30 12:27:14',
 '2016-01-30 12:27:36']

但是，如果您想对与这些时间戳关联的数据重新采样，最好将其保留为 pd.Series 并跳过最后一步。最后你可能想使用 pd.DataFrame.resample无论如何。