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 !