我已经看到熊猫无声地排除了滋扰栏,如下所述:Pandas Nuisance columns
它声称如果无法将聚合函数应用于列,它会以静默方式排除列。
考虑以下示例:
我有一个数据框:
df = pd.DataFrame({'C': {0: -0.91985400000000006, 1: -0.042379, 2: 1.2476419999999999, 3: -0.00992, 4: 0.290213, 5: 0.49576700000000001, 6: 0.36294899999999997, 7: 1.548106}, 'A': {0: 'foo', 1: 'bar', 2: 'foo', 3: 'bar', 4: 'foo', 5: 'bar', 6: 'foo', 7: 'foo'}, 'B': {0: -1.131345, 1: -0.089328999999999992, 2: 0.33786300000000002, 3: -0.94586700000000001, 4: -0.93213199999999996, 5: 1.9560299999999999, 6: 0.017587000000000002, 7: -0.016691999999999999}})
df:
A B C
0 foo -1.131345 -0.919854
1 bar -0.089329 -0.042379
2 foo 0.337863 1.247642
3 bar -0.945867 -0.009920
4 foo -0.932132 0.290213
5 bar 1.956030 0.495767
6 foo 0.017587 0.362949
7 foo -0.016692 1.548106
让我将两列B和C组合起来并转换为numpy ndarray:
df = df.assign(D=df[['B', 'C']].values.tolist())
df['D'] = df['D'].apply(np.array)
df:
A B C D
0 foo -1.131345 -0.919854 [-1.131345, -0.9198540000000001]
1 bar -0.089329 -0.042379 [-0.08932899999999999, -0.042379]
2 foo 0.337863 1.247642 [0.337863, 1.247642]
3 bar -0.945867 -0.009920 [-0.945867, -0.00992]
4 foo -0.932132 0.290213 [-0.932132, 0.290213]
5 bar 1.956030 0.495767 [1.95603, 0.495767]
6 foo 0.017587 0.362949 [0.017587000000000002, 0.36294899999999997]
7 foo -0.016692 1.548106 [-0.016692, 1.548106]
现在我可以将均值应用于D列:
print(df['D'].mean())
print(df['B'].mean())
print(df['C'].mean())
[-0.10048563 0.3715655 ]
-0.100485625
0.3715655
但是当我尝试用A组合并获得平均值时,D列就会被删除:
df.groupby('A').mean()
B C
A
bar 0.306945 0.147823
foo -0.344944 0.505811
我的问题是,为什么D列被排除在外,即使可以成功应用聚合函数?
而且,一般来说,当一个特定的感兴趣的列是一个numpy数组时,我如何使用像mean或sum这样的聚合函数?
答案 0 :(得分:0)
是否可行,但在自定义函数中需要if-else
:
def f(x):
a = x.mean()
return a if isinstance(a, (float, int)) else list(a)
df = df.groupby('A').agg(f)
print (df)
B C D
A
bar 0.306945 0.147823 [0.306944666667, 0.147822666667]
foo -0.344944 0.505811 [-0.3449438, 0.5058112]
df = df.groupby('A').agg(lambda x: x.mean())
print (df)
B C D
A
bar 0.306945 0.147823 NaN
foo -0.344944 0.505811 NaN