CODEWITHSUNDEEP
numpymultidimensional-arrayshuffle
AutomEng
AutomEng

Reputation: 455

Shuffle groups of rows of a 2D array - NumPy

Suppose I have a (50, 5) array. Is there a way for me to shuffle it on the basis of groupings of rows/sequences of datapoints, i.e. instead of shuffling every row, shuffle chunks of say, 5 rows?

Thanks

Upvotes: 2

Views: 744

Answers (1)

Divakar
Divakar

Reputation: 221524

Approach #1 : Here's an approach that reshapes into a 3D array based on the group size, indexes into the indices of blocks with shuffled indices obtained from np.random.permutation and finally reshapes back to 2D -

N = 5 # Blocks of N rows
M,n = a.shape[0]//N, a.shape[1]
out = a.reshape(M,-1,n)[np.random.permutation(M)].reshape(-1,n)

Sample run -

In [141]: a
Out[141]: 
array([[89, 26, 12],
       [97, 60, 96],
       [94, 38, 54],
       [41, 63, 29],
       [88, 62, 48],
       [95, 66, 32],
       [28, 58, 80],
       [26, 35, 89],
       [72, 91, 38],
       [26, 70, 93]])

In [142]: N = 2 # Blocks of N rows

In [143]: M,n = a.shape[0]//N, a.shape[1]

In [144]: a.reshape(M,-1,n)[np.random.permutation(M)].reshape(-1,n)
Out[144]: 
array([[94, 38, 54],
       [41, 63, 29],
       [28, 58, 80],
       [26, 35, 89],
       [89, 26, 12],
       [97, 60, 96],
       [72, 91, 38],
       [26, 70, 93],
       [88, 62, 48],
       [95, 66, 32]])

Approach #2 : One can also simply use np.random.shuffle for an in-situ change -

np.random.shuffle(a.reshape(M,-1,n))

Sample run -

In [156]: a
Out[156]: 
array([[15, 12, 14],
       [55, 39, 35],
       [73, 78, 36],
       [54, 52, 32],
       [83, 34, 91],
       [42, 11, 98],
       [27, 65, 47],
       [78, 75, 82],
       [33, 52, 93],
       [87, 51, 80]])

In [157]: N = 2 # Blocks of N rows

In [158]: M,n = a.shape[0]//N, a.shape[1]

In [159]: np.random.shuffle(a.reshape(M,-1,n))

In [160]: a
Out[160]: 
array([[15, 12, 14],
       [55, 39, 35],
       [27, 65, 47],
       [78, 75, 82],
       [73, 78, 36],
       [54, 52, 32],
       [33, 52, 93],
       [87, 51, 80],
       [83, 34, 91],
       [42, 11, 98]])

Upvotes: 3

Related Questions