How to shuffle groups of rows of a Pandas dataframe?

Question

Let's assume I have a dataframe df:

import numpy as np
import pandas as pd

df = pd.DataFrame(np.random.rand(12,4))

print(df)

     0   1   2   3
0   71  64  84  20
1   48  60  83  61
2   48  78  71  46
3   65  88  66  77
4   71  22  42  58
5   66  76  64  80
6   67  28  74  87
7   32  90  55  78
8   80  42  52  14
9   54  76  73  17
10  32  89  42  36
11  85  78  61  12

How do I shuffle the rows of df three-by-three, i.e., how do I randomly shuffle the first three rows (0, 1, 2) with either the second (3, 4, 5), third (6, 7, 8) or fourth (9, 10, 11) group? This could be a possible outcome:

print(df)

     0   1   2   3
3   65  88  66  77
4   71  22  42  58
5   66  76  64  80
9   54  76  73  17
10  32  89  42  36
11  85  78  61  12
6   67  28  74  87
7   32  90  55  78
8   80  42  52  14
0   71  64  84  20
1   48  60  83  61
2   48  78  71  46

Thus, the new order has the second group of 3 rows from original dataframe, then the last one, then the third one and finally the first group.

Answer 1

You can reshape into a 3D array splitting the first axis into two with the latter one of length 3 corresponding to the group length and then use np.random.shuffle for such a groupwise in-place shuffle along the first axis, which being of length as the number of groups holds those groups and thus achieves our desired result, like so -

np.random.shuffle(df.values.reshape(-1,3,df.shape[1]))

Explanation

To give it a bit of explanation, let's use np.random.permutation to generate those random indices along the first axis and then index into the 3D array version.

1] Input df :

In [199]: df
Out[199]: 
     0   1   2   3
0   71  64  84  20
1   48  60  83  61
2   48  78  71  46
3   65  88  66  77
4   71  22  42  58
5   66  76  64  80
6   67  28  74  87
7   32  90  55  78
8   80  42  52  14
9   54  76  73  17
10  32  89  42  36
11  85  78  61  12

2] Get 3D array version :

In [200]: arr_3D = df.values.reshape(-1,3,df.shape[1])

In [201]: arr_3D
Out[201]: 
array([[[71, 64, 84, 20],
        [48, 60, 83, 61],
        [48, 78, 71, 46]],

       [[65, 88, 66, 77],
        [71, 22, 42, 58],
        [66, 76, 64, 80]],

       [[67, 28, 74, 87],
        [32, 90, 55, 78],
        [80, 42, 52, 14]],

       [[54, 76, 73, 17],
        [32, 89, 42, 36],
        [85, 78, 61, 12]]])

3] Get shuffling indices and index into the first axis of 3D version :

In [202]: shuffle_idx = np.random.permutation(arr_3D.shape[0])

In [203]: shuffle_idx
Out[203]: array([0, 3, 1, 2])

In [204]: arr_3D[shuffle_idx]
Out[204]: 
array([[[71, 64, 84, 20],
        [48, 60, 83, 61],
        [48, 78, 71, 46]],

       [[54, 76, 73, 17],
        [32, 89, 42, 36],
        [85, 78, 61, 12]],

       [[65, 88, 66, 77],
        [71, 22, 42, 58],
        [66, 76, 64, 80]],

       [[67, 28, 74, 87],
        [32, 90, 55, 78],
        [80, 42, 52, 14]]])

Then, we are assigning these values back to input dataframe.

With np.random.shuffle, we are just doing everything in-place and hiding away the work needed to explicitly generate shuffling indices and assigning back.

Sample run -

In [181]: df = pd.DataFrame(np.random.randint(11,99,(12,4)))

In [182]: df
Out[182]: 
     0   1   2   3
0   82  49  80  20
1   19  97  74  81
2   62  20  97  19
3   36  31  14  41
4   27  86  28  58
5   38  68  24  83
6   85  11  25  88
7   21  31  53  19
8   38  45  14  72
9   74  63  40  94
10  69  85  53  81
11  97  96  28  29

In [183]: np.random.shuffle(df.values.reshape(-1,3,df.shape[1]))

In [184]: df
Out[184]: 
     0   1   2   3
0   85  11  25  88
1   21  31  53  19
2   38  45  14  72
3   82  49  80  20
4   19  97  74  81
5   62  20  97  19
6   36  31  14  41
7   27  86  28  58
8   38  68  24  83
9   74  63  40  94
10  69  85  53  81
11  97  96  28  29

Answer 2

Similar solution to @Divakar, probably simpler as I directly shuffle the index of the dataframe:

import numpy as np
import pandas as pd

df = pd.DataFrame([np.arange(0, 12)]*4).T
len_group = 3

index_list = np.array(df.index)
np.random.shuffle(np.reshape(index_list, (-1, len_group)))

shuffled_df = df.loc[index_list, :]

Sample output:

shuffled_df
    Out[82]: 
     0   1   2   3
9    9   9   9   9
10  10  10  10  10
11  11  11  11  11
3    3   3   3   3
4    4   4   4   4
5    5   5   5   5
0    0   0   0   0
1    1   1   1   1
2    2   2   2   2
6    6   6   6   6
7    7   7   7   7
8    8   8   8   8