Question

我希望做到以下几点：

对数据框进行分组
对于每个组，生成时间窗口（给定时间单位）
在结果结构中，取每列并应用多个滚动摘要统计函数，以便结果具有每个组/时间窗组合的摘要统计量。

以下是一个示例数据集：

gps_time,name,val_x,val_y
2017-07-04 11:20:23.423,bob,0.963,0.201
2017-07-04 11:20:24.492,bob,0.964,0.203
2017-07-04 11:20:24.499,bob,0.962,0.210
2017-07-04 11:20:25.627,sarah,0.893,0.010
2017-07-04 11:20:28.627,sarah,0.894,0.012
2017-07-04 11:20:29.613,sarah,0.895,0.014
2017-07-04 11:20:29.630,larry,-0.423,0.231
2017-07-04 11:20:30.423,larry,-0.431,0.22
2017-07-04 11:20:30.428,larry,-0.432,0.222

以上数据的所需输出，按名称分组，窗口为1秒：

name,gps_time,val_x_mean,val_x_med,val_y_mean,val_y_med
bob,2017-07-04 11:20:23.423,0.963,0.963,0.201,0.201
bob,2017-07-04 11:20:24.492,0.963,0.963,0.2065,0.2065
sarah,2017-07-04 11:20:25.627,0.893,0.89,0.010,0.010
sarah,2017-07-04 11:20:28.627,0.8945,0.8945,0.013,0.013
larry,2017-07-04 11:20:30.423,-0.4287,-0.431,0.336,0.222

我尝试使用列表推导来生成一堆数据帧，但这个过程非常慢，我必须为每一列调用它。

Answer 1

我们将groupby与pd.Grouper：

一起使用

df_out = df.groupby([pd.Grouper(freq='S', key='gps_time'),'name']).agg(['mean','median'])
df_out.columns = df_out.columns.map('_'.join)
df_out.reset_index()

输出：

             gps_time   name  val_x_mean  val_x_median  val_y_mean  \
0 2017-07-04 11:20:23    bob      0.9630        0.9630      0.2010   
1 2017-07-04 11:20:24    bob      0.9630        0.9630      0.2065   
2 2017-07-04 11:20:25  sarah      0.8930        0.8930      0.0100   
3 2017-07-04 11:20:28  sarah      0.8940        0.8940      0.0120   
4 2017-07-04 11:20:29  larry     -0.4230       -0.4230      0.2310   
5 2017-07-04 11:20:29  sarah      0.8950        0.8950      0.0140   
6 2017-07-04 11:20:30  larry     -0.4315       -0.4315      0.2210   

   val_y_median  
0        0.2010  
1        0.2065  
2        0.0100  
3        0.0120  
4        0.2310  
5        0.0140  
6        0.2210

通过滚动对象将多个滚动函数应用于pandas组的多个列？

1 个答案: