将 DataFrame 转换为多维时间序列？答案

【问题标题】：Transform DataFrame into multidimensional TimeSeries?将 DataFrame 转换为多维时间序列？
【发布时间】：2021-08-08 13:03:04
【问题描述】：

我在 'county' 列上有以下 pandas DataFrame 的“周期性”值，以及 'reporting_period' 和 'date' 中的重复值：

data = pd.DataFrame({'county': {0: 'A', 1: 'B', 2: 'C', 3: 'D', 4: 'E', 5: 'F', 6: 'G', 7: 'H', 8: 'I', 9: 'A', 10: 'B', 11: 'C', 12: 'D', 13: 'E', 14: 'F', 15: 'G', 16: 'H', 17: 'I'}, 'new_covid_19_cases_per_100k': {0: 9.89857311398793, 1: 8.96808587445497, 2: 10.4018656786281, 3: 5.44259755461725, 4: 8.47402557487262, 5: 8.23708135804402, 6: 21.1781816000959, 7: 6.34201242466493, 8: 11.9630512616746, 9: 14.0, 10: 16.3, 11: 13.1, 12: 9.3, 13: 11.0, 14: 12.6, 15: 20.9, 16: 8.2, 17: 13.6}, 'new_covid_19_hospitalizations': {0: 0.735745284982339, 1: 0.681120446161137, 2: 1.07219230841243, 3: 0.118317338143853, 4: 0.526882419163064, 5: 0.599666185823225, 6: 1.07095735019448, 7: 0.141985352791006, 8: 0.854503661548189, 9: 0.9, 10: 0.8, 11: 1.5, 12: 0.2, 13: 0.5, 14: 0.8, 15: 0.9, 16: 0.1, 17: 0.7}, 'reporting_period': {0: '10/04/2020 - 10/17/2020', 1: '10/04/2020 - 10/17/2020', 2: '10/04/2020 - 10/17/2020', 3: '10/04/2020 - 10/17/2020', 4: '10/04/2020 - 10/17/2020', 5: '10/04/2020 - 10/17/2020', 6: '10/04/2020 - 10/17/2020', 7: '10/04/2020 - 10/17/2020', 8: '10/04/2020 - 10/17/2020', 9: '10/11/2020 - 10/24/2020', 10: '10/11/2020 - 10/24/2020', 11: '10/11/2020 - 10/24/2020', 12: '10/11/2020 - 10/24/2020', 13: '10/11/2020 - 10/24/2020', 14: '10/11/2020 - 10/24/2020', 15: '10/11/2020 - 10/24/2020', 16: '10/11/2020 - 10/24/2020', 17: '10/11/2020 - 10/24/2020'}, 'date': {0: '2020-10-22T00:00:00', 1: '2020-10-22T00:00:00', 2: '2020-10-22T00:00:00', 3: '2020-10-22T00:00:00', 4: '2020-10-22T00:00:00', 5: '2020-10-22T00:00:00', 6: '2020-10-22T00:00:00', 7: '2020-10-22T00:00:00', 8: '2020-10-22T00:00:00', 9: '2020-10-29T00:00:00', 10: '2020-10-29T00:00:00', 11: '2020-10-29T00:00:00', 12: '2020-10-29T00:00:00', 13: '2020-10-29T00:00:00', 14: '2020-10-29T00:00:00', 15: '2020-10-29T00:00:00', 16: '2020-10-29T00:00:00', 17: '2020-10-29T00:00:00'}})

我的目标是将此 DataFrame 转换为某种多维时间序列，但我不知道最好的方法是什么，也不知道这是否可能。

我的第一个想法是使用groupby 和pivot_table，但我不确定这是否有用。

【问题讨论】：

你至少知道预期的结果是什么吗？

标签： python pandas datetime time-series

【解决方案1】：

将 ts 数据可视化为multiindex 的最简单方法是set_index。 reporting_period 也可以转换为 period type 但这取决于要求。
如果我们想申请任何aggregation, reduction 或任何其他transformation，那么我们将不得不使用groupby or pivot。

data['date'] = pd.to_datetime(data.date)
data = data.set_index(['reporting_period', 'date'])
data

样本输出

data.head(2)

                                    county  new_covid_19_cases_per_100k new_covid_19_hospitalizations
reporting_period        date            
10/04/2020 - 10/17/2020 2020-10-22  A       9.898573    0.735745
                        2020-10-22  B       8.968086    0.681120

【讨论】：