Numpy; reshape 1d number sequence to 2d number sequences-CodePudding

Soppose I have a numpy array like this:

arr = np.array([0,1,2,5,6,7,10,11,12,15,16,17])

And I want the following result:

array([[ 0,  1,  2],
       [ 5,  6,  7],
       [10, 11, 12]
       [15, 16, 17]])

Another example would be:

arr = np.array([3900, 3901, 3902, 3903, 3964, 3965, 3966, 3967, 4028, 4029, 4030, 4031, 4092, 4093, 4094, 4095])

Result:

array([[3900, 3901, 3902, 3903],
       [3964, 3965, 3966, 3967],
       [4028, 4029, 4030, 4031],
       [4092, 4093, 4094, 4095]])

These 1d arrays are generated dynamically with different length.

result = arr.reshape((?,?))

The resulting array must consist of rows that are in sequential order. When the sequence breaks, that's a new row.

CodePudding user response：

Assuming consecutive values with breaks at regular intervals, like in your two examples. You can use np.diff(arr) to compute the difference between successive values and np.argmax(np.diff(arr)!=1) to identify the index of the first "break" in the increasing sequence. The width of the reshape is this number 1.

We can now use reshape:

arr = np.array([3900, 3901, 3902, 3903, 3964, 3965, 3966, 3967, 4028, 4029, 4030, 4031, 4092, 4093, 4094, 4095])

arr.reshape((-1, (1 np.argmax(np.diff(arr)!=1))))

output:

array([[3900, 3901, 3902, 3903],
       [3964, 3965, 3966, 3967],
       [4028, 4029, 4030, 4031],
       [4092, 4093, 4094, 4095]])

CodePudding user response：

The following works with both arrays, but would require that the rows have all the same length.

# get indices where sequence breaks
indices = np.where(np.diff(arr)!=1)[0]

# get lengths of the sequences (except one is missing)
row_length = np.diff(indices)

# check if row length matches
if np.all(row_length == row_length[0]):
    reshaped = arr.reshape((-1, row_length[0]))
    print(arr)
    print(reshaped)
else:
    print('unable to reshape, sequences have different lengths')

Output:

[ 0  1  2  5  6  7 10 11 12 15 16 17]
[[ 0  1  2]
 [ 5  6  7]
 [10 11 12]
 [15 16 17]]

[3900 3901 3902 3903 3964 3965 3966 3967 4028 4029 4030 4031 4092 4093
 4094 4095]
[[3900 3901 3902 3903]
 [3964 3965 3966 3967]
 [4028 4029 4030 4031]
 [4092 4093 4094 4095]]

CodePudding user response：

import numpy as np

arr = np.array([0,1,2,5,6,7,10,11,12,15,16,17])

result = arr.reshape(4, -1)

print(result)

Output:

[[ 0,  1,  2],
[ 5,  6,  7],
[10, 11, 12],
[15, 16, 17]]