Question

我正在尝试根据其周数来计算数据框特定子集的变化百分比。数据框如下所示：

      ref_dt   week_name  county_name  state_name   county_fips_code    cmi
0   2020-01-01  2020-W01   Broward      Florida         12011         3.651278
1   2020-01-02  2020-W01.  Broward      Florida         12011         3.851842
2   2020-01-03  2020-W01.  Broward      Florida         12011         3.868523
3   2020-01-04  2020-W01.  Broward      Florida         12011         3.748446
4   2020-01-05  2020-W01.  Broward      Florida         12011         3.650769
5   2020-01-06  2020-W02.  Broward      Florida         12011         3.878860
6   2020-01-07  2020-W02.  Broward      Florida         12011         3.899171
7   2020-01-08  2020-W02.  Broward      Florida         12011         3.907816
8   2020-01-09  2020-W02.  Broward      Florida         12011         3.913623
9   2020-01-10  2020-W02.  Broward      Florida         12011         3.919010

它具有佛罗里达州每个县的信息（此处仅以布劳沃德为例），并且在cmi列中计算了流动性指数。通过将一周中某天的移动量（ref_dt）与一周中同一天的平均值进行比较来计算百分比变化。这是我在熊猫中进行次选择后的第1周和Broward的示例。

df = counties[counties['county_name']=='Broward']
week1 = df[df['week_name'] == '2020-W01']
cmi_mean = week1['cmi'].mean()
week1['percent_change'] = week1['cmi']/cmi_mean * 100

csv的最终输出看起来像这样（我删除了state和county_fips_code）：

      ref_dt    week_name  county_name    cmi      percent_change
0   2020-01-01  2020-W01     Broward    3.651278    97.259216
1   2020-01-02  2020-W01     Broward    3.851842    102.601642
2   2020-01-03  2020-W01     Broward    3.868523    103.045989
3   2020-01-04  2020-W01     Broward    3.748446    99.847497
4   2020-01-05  2020-W01     Broward    3.650769    97.245656

我想为每个县（每周1到14）应用相同的逻辑。最好的方法是什么？我是否需要使用数据透视表或堆栈重塑数据框并根据它们的week_name每周创建列，还是可以计算数据框当前结构的变化百分比？

注意：必须每周计算每个平均值。

Answer 1

将df.groupby与transform一起使用，并让熊猫使用索引来处理计算对齐：

df['percent_change'] = df['cmi'] / df.groupby(['county_name', 'week_name'])['cmi'].transform('mean') * 100

输出：

       ref_dt  week_name county_name state_name  county_fips_code       cmi  percent_change
0  2020-01-01  2020-W01.     Broward    Florida             12011  3.651278       97.259220
1  2020-01-02  2020-W01.     Broward    Florida             12011  3.851842      102.601650
2  2020-01-03  2020-W01.     Broward    Florida             12011  3.868523      103.045982
3  2020-01-04  2020-W01.     Broward    Florida             12011  3.748446       99.847487
4  2020-01-05  2020-W01.     Broward    Florida             12011  3.650769       97.245661
5  2020-01-06  2020-W02.     Broward    Florida             12011  3.878860       99.363782
6  2020-01-07  2020-W02.     Broward    Florida             12011  3.899171       99.884084
7  2020-01-08  2020-W02.     Broward    Florida             12011  3.907816      100.105541
8  2020-01-09  2020-W02.     Broward    Florida             12011  3.913623      100.254297
9  2020-01-10  2020-W02.     Broward    Florida             12011  3.919010      100.392295

如何对熊猫中数据帧的子组执行操作？

1 个答案: