Getting TypeError: unhashable type: 'slice' when slicing multilevel column using loc-CodePudding

The objective is to slice level 0 with a in it.

However, when slicing the multiindex column as below,

a_cols=df.loc[:,('a',slice(None))]

the compiler return

TypeError: unhashable type: 'slice'

import pandas as pd
import numpy as np
np.random.seed(0)
arr=np.random.randint(5, size=(2, 12))

df=pd.DataFrame(arr,columns=[('a','E1_g1'),('a','E1_g2'),('a','E1_g3'),
                                                        ('a','E2_g1'),('a','E2_g2'),('a','E2_g3'),
                                                        ('a','E3_g1'),('a','E3_g2'),('a','E3_g3'),
                                                        ('b','E1'),('b','E1'),('b','E13'),])

May I know what did I do wrong here?

I also tried

df.loc[:, df.columns.get_level_values(0) == 'a']

But, an empty df is produced instead.

CodePudding user response：

That's because df.columns is not MultiIndex. You can index it using a list created by filtering df.columns:

cols = [(i,j) for (i,j) in df.columns if i=='a']
out = df[cols]

Output:

   (a, E1_g1)  (a, E1_g2)  (a, E1_g3)  (a, E2_g1)  (a, E2_g2)  (a, E2_g3)  (a, E3_g1)  (a, E3_g2)  (a, E3_g3)
0           4           0           3           3           3           1           3           2           4
1           2           1           0           1           1           0           1           4           3

You can make df.columns MultiIndex using MultiIndex.from_tuples. Then your slicing method works:

df.columns = pd.MultiIndex.from_tuples(df.columns)
a_cols = df.loc[:,('a',slice(None))]

Output:

      a                                                
  E1_g1 E1_g2 E1_g3 E2_g1 E2_g2 E2_g3 E3_g1 E3_g2 E3_g3
0     4     0     3     3     3     1     3     2     4
1     2     1     0     1     1     0     1     4     3