How to bin a 2D array in numpy?

You can reshape the array to a four dimensional array that reflects the desired block structure, and then sum along both axes within each block. Example:

>>> a = np.arange(24).reshape(4, 6)>>> aarray([[ 0,  1,  2,  3,  4,  5],       [ 6,  7,  8,  9, 10, 11],       [12, 13, 14, 15, 16, 17],       [18, 19, 20, 21, 22, 23]])>>> a.reshape(2, 2, 2, 3).sum(3).sum(1)array([[ 24,  42],       [ 96, 114]])

If a has the shape m, n, the reshape should have the form

a.reshape(m_bins, m // m_bins, n_bins, n // n_bins)

At first I was also going to suggest that you use np.histogram2d rather than reinventing the wheel, but then I realized that it would be overkill to use that and would need some hacking still.

If I understand correctly, you just want to sum over submatrices of your input. That's pretty easy to brute force: going over your output submatrix and summing up each subblock of your input:

import numpy as npdef submatsum(data,n,m):    # return a matrix of shape (n,m)    bs = data.shape[0]//n,data.shape[1]//m  # blocksize averaged over    return np.reshape(np.array([np.sum(data[k1*bs[0]:(k1+1)*bs[0],k2*bs[1]:(k2+1)*bs[1]]) for k1 in range(n) for k2 in range(m)]),(n,m))# set up dummy dataN,M = 4,6data_matrix = np.reshape(np.arange(N*M),(N,M))# set up size of 2x3-reduced matrix, assume congruityn,m = N//2,M//3reduced_matrix = submatsum(data_matrix,n,m)# check outputprint(data_matrix)print(reduced_matrix)

This prints

print(data_matrix)[[ 0  1  2  3  4  5] [ 6  7  8  9 10 11] [12 13 14 15 16 17] [18 19 20 21 22 23]]print(reduced_matrix)[[ 24  42] [ 96 114]]

which is indeed the result for summing up submatrices of shape (2,3).

Note that I'm using // for integer division to make sure it's python3-compatible, but in case of python2 you can just use / for division (due to the numbers involved being integers).

Another solution is to have a look at the binArray function on the comments here: Binning a numpy array

To use your example :

data_matrix = numpy.ndarray((500,500),dtype=float)binned_data = binArray(data_matrix, 0, 10, 10, np.sum)binned_data = binArray(binned_data, 1, 10, 10, np.sum)

The result sum all square of size 10x10 in data_matrix (of size 500x500) to obtain a single value per square in binned_data (of size 50x50).

Hope this help !