[FIXED] How to properly loop with Numpy array of tuples?

Issue

I have a Numpy 2D matrix of zeros, and a Numpy list of coordinates. What I would like to do, is modify the matrix thanks to the list of coordinates by passing it in a function.

Here is what I tried :

import numpy as np

def update(coords):
    return np_arr[coords[0]][coords[1]]+1

size = 3
np_arr = np.zeros((size,size))
dt = np.dtype('int', 'int')
np_indices = np.array([(x,y) for y in range(size) for x in range(size)], dtype=dt)
np_arr = update(np_indices)
print(np_arr)

I expected the output to be a size*size matrix:

[[1. 1. 1.]
 [1. 1. 1.]
 [1. 1. 1.]]

But instead I get:

[[1. 1. 1.]
 [1. 1. 1.]]

Why does the shape of the matrix is not the same?

I know that I could simply use:

np_arr = np.array([[update(x,y) for y in range(size)] for x in range(size)])

However, I would like to avoid using np.array every time.

Solution

In [484]: size = 3
     ...: np_arr = np.zeros((size,size))
     ...: dt = np.dtype('int', 'int')
     ...: np_indices = np.array([(x,y) for y in range(size) for x in range(size)], dtype=dt)

In [485]: np_arr
Out[485]: 
array([[0., 0., 0.],
       [0., 0., 0.],
       [0., 0., 0.]])

Despite using the compound dt, the np_indices array is a 2d int dtype array. Were you hoping to get an array of tuples?

In [486]: np_indices
Out[486]: 
array([[0, 0],
       [1, 0],
       [2, 0],
       [0, 1],
       [1, 1],
       [2, 1],
       [0, 2],
       [1, 2],
       [2, 2]])

Anyways, the way to index np_arr is to use the 2 columns, as in:

In [487]: np_arr[np_indices[:,0],np_indices[:,1]]
Out[487]: array([0., 0., 0., 0., 0., 0., 0., 0., 0.])

In [488]: np_arr[np_indices[:,0],np_indices[:,1]]+=1

In [489]: np_arr
Out[489]: 
array([[1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.]])

You can't use np_arr[np_indices[:,0]][np_indices[:,1]. That applies the 2 indexing operations sequentially. It works for scalar indices, np_arr[2,3], np_arr[2][3], but not when using arrays. And np_indices[0] is the first row, not column.

In [490]: size = 3
     ...: np_arr = np.zeros((size,size))
     ...: dt = np.dtype([('x','int'), ('y','int')])
     ...: np_indices = np.array([(x,y) for y in range(size) for x in range(size)], dtype=dt)

In [491]: np_indices
Out[491]: 
array([(0, 0), (1, 0), (2, 0), (0, 1), (1, 1), (2, 1), (0, 2), (1, 2),
       (2, 2)], dtype=[('x', '<i4'), ('y', '<i4')])

With a different dt, we get a 1d structured array:

In [492]: np_indices['x']
Out[492]: array([0, 1, 2, 0, 1, 2, 0, 1, 2])

We still need to use the 2 array indexing:

In [493]: np_arr[np_indices['x'],np_indices['y']]
Out[493]: array([0., 0., 0., 0., 0., 0., 0., 0., 0.])

The structured array could be used in loop:

In [495]: [np_arr[x,y] for x,y in np_indices]
Out[495]: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]

but generally we avoid that.

Answered By - hpaulj