Populating a NumPy array using dictionary containing indices

Question

I am trying to populate a numpy array based on a dictionary of values.

My dictionary looks like this:

A = {(12, 15): 4, (532, 31): 7, (742, 1757): 1, ...}

I am trying to populate the array so that (using my above example) 4 is at the index (12,15) and so on. The keys in A are called 'd','s' and the value is referred to as 'count'.

A = {(d,s): count}

At the moment my code to populate the array looks like this:

N = len(seen)
Am = np.zeros((N,N), 'i')
for key, count in A.items():
    Am[d,s] = count

But this just leads to multiple arrays mostly full of zeros being created.

Thank you

Answer 1

Here's one approach -

def dict_to_arr(A):
    idx = np.array(list(A.keys()))
    val = np.array(list(A.values()))

    m,n = idx.max(0)+1 # max extents of indices to decide o/p array
    out = np.zeros((m,n), dtype=val.dtype)
    out[idx[:,0], idx[:,1]] = val # or out[tuple(idx.T)] = val
    return out

Possibly faster one if we avoid the array conversion of the indices and values and directly use those for assigning at the last step, like so -

out[zip(*A.keys())] = list(A.values())

Sample run -

In [3]: A = {(12, 15): 4, (532, 31): 7, (742, 1757): 1}

In [4]: arr = dict_to_arr(A)

In [5]: arr[12,15], arr[532,31], arr[742,1757]
Out[5]: (4, 7, 1)

---

Store into sparse matrix

To save on memory and possibly gain performance as well, we might want to store in a sparse matrix. Let's do it with csr_matrix, like so -

from scipy.sparse import csr_matrix

def dict_to_sparsemat(A):
    idx = np.array(list(A.keys()))
    val = np.array(list(A.values()))
    m,n = idx.max(0)+1
    return csr_matrix((val, (idx[:,0], idx[:,1])), shape=(m,n))

Sample run -

In [64]: A = {(12, 15): 4, (532, 31): 7, (742, 1757): 1}

In [65]: out = dict_to_sparsemat(A)

In [66]: out[12,15], out[532,31], out[742,1757]
Out[66]: (4, 7, 1)