Python:从长格式 Pandas 数据帧创建嵌套列表

Question

我有一个数据框，其中只有“peak_time”是一列:

stimulus position peak_time 
1        1        1.0
         2        1.5
2        1        2.0
         2        2.0
3        1        2.5

现在我试图压缩位置列并获取列表，因此它看起来像这样:

stimulus peak_time  
1        [1.0, 1.5]
2        [2.0, 2.0]
3        [2.5]

这可能非常简单，但我找不到任何使用 goole 的解决方案。如果有人已经打开了这个主题，我也会很感激相应的链接。感谢您的帮助!

创建数据框的代码:

import random, scipy
import pandas as pd
trial     = [1,1,2,1,1,2,2,1,2]
stimulus  = [1,1,1,2,2,2,2,3,3] 
position  = [1,2,1,1,2,1,2,1,1]
peak_time = random.sample(range(1000), 9)
df        = pd.DataFrame({"trial": trial, "stimulus": stimulus, "position": position, "peak_time": peak_time})
median_   = df.groupby(['stimulus', 'position']).apply(scipy.nanmedian)
median_   = pd.DataFrame(median_)
median_.columns = ['peak_time']
median_

编辑

由于我每 90 分钟只能发布一个问题，所以我想在这篇文章下面提出一个跟进问题。所以现在我有两个像这样的 Pandas 系列:

median_:
stimulus
1    [1.0, 1.5]
2    [2.0, 2.0]
3    [2.0]

quartile_:
stimulus
1    [[1.0, 70.0],  [1.0, 183.25]]
2    [[1.0, 65.75], [2.0, 98.75]]
3    [[1.0, 51.25]]

我想从 quartile_ 中减去 median_ 得到

distance_: 
stimulus
1   [1-1, 70-1], [1.5-1, 183.25-1.5]
2   [2-1, 65.75-1], [2-2, 98.75-2]
3   [2-1, 51.25-2]

有没有简单的方法来做到这一点？ abs(median_ - quartile_) 不起作用。

创建系列的代码:

import random, scipy
import pandas as pd
trial     = [1,1,2,1,1,2,2,1,2]
stimulus  = [1,1,1,2,2,2,2,3,3] 
position  = [1,2,1,1,2,1,2,1,1]
peak_time = random.sample(range(1000), 9)
df        = pd.DataFrame({"trial": trial, "stimulus": stimulus, "position": position, "peak_time": peak_time})
median_   = df.groupby(['stimulus', 'position']).apply(scipy.nanmedian).groupby(level=0).apply(list)
quartile_ = df.groupby(['stimulus', 'position']).apply(lambda x: scipy.nanpercentile(x, [25, 75])).groupby(level=0).apply(list)

解决方案

稍后应用groupby(level=0).apply(list)，所以

median_   = df.groupby(['stimulus', 'position']).apply(scipy.nanmedian)
quartile_ = df.groupby(['stimulus', 'position']).apply(lambda x: scipy.nanpercentile(x, [25, 75]))

然后我可以很容易地减去它们

distance_ = abs(median_ - quartile_)
distance_ = distance.groupby(level=0).apply(list)
distance_

stimulus
1   [1-1, 70-1], [1.5-1, 183.25-1.5]
2   [2-1, 65.75-1], [2-2, 98.75-2]
3   [2-1, 51.25-2]

Answer 1

它是MultiIndex Series，所以需要Series.groupby与 apply 列表:

#added column peak_time
median_   = df.groupby(['stimulus', 'position'])['peak_time'].apply(scipy.nanmedian)
df        = median_.groupby(level=0).apply(list).reset_index()
print (df)
   stimulus   peak_time
0         1  [1.0, 1.5]
1         2  [2.0, 2.0]
2         3       [2.5]