Euclidean Distance Calculation Vectorization - NumPy

Question

my question is about vectorizing my code. I have one array that contains 3D coordinates and one array that contains information about the edges that connect the coordinates:

In [8]:coords
Out[8]: 
array([[ 11.22727013,  24.72620964,   2.02986932],
       [ 11.23895836,  24.67577744,   2.04130101],
       [ 11.23624039,  24.63677788,   2.04096866],
       [ 11.22516632,  24.5986824 ,   2.04045677],
       [ 11.21166992,  24.56095695,   2.03898215],
       [ 11.20334721,  24.5227356 ,   2.03556442],
       [ 11.2064085 ,  24.48479462,   2.03098583],
       [ 11.22059727,  24.44837189,   2.02649784],
       [ 11.24213409,  24.41513252,   2.01979685]])

In [13]:edges
Out[13]: 
array([[0, 1],
       [1, 2],
       [2, 3],
       [3, 4],
       [4, 5],
       [5, 6],
       [6, 7],
       [7, 8],], dtype=int32)

Now I would like to calculate the sum of the Euclidean distance between the coordinates in the array of edges. For instance. Distance from coords [0] to coords [1] + distance from coords [1] to coords [2] .....

I have the following code that does the job:

def networkLength(coords, edges):

   from scipy.spatial import distance 
   distancesNetwork = np.array([])    

   for i in range(edges.shape[0]):
        distancesNetwork = np.append(distancesNetwork, distance.euclidean(coords[edges[i, 0]], coords[edges[i, 1]]))

   return sum(distancesNetwork)

I was wondering if it is possible to vectorize the code, rather than looping. What is the Python way to do this? Many thanks!

Answer 1

№ 1

coords , , - . , , .

, -

np.sqrt(((coords[edges[:, 0]] - coords[edges[:, 1]])**2).sum(1)).sum()

NumPy NumPy np.linalg.norm. , , , . -

np.linalg.norm(coords[edges[:, 0]] - coords[edges[:, 1]],axis=1).sum()

# 2

, np.einsum, squaring summing along each row, .

:

s = coords[edges[:, 0]] - coords[edges[:, 1]]
out = np.sqrt(np.einsum('ij,ij->i',s,s)).sum()

---

-

def networkLength(coords, edges): # Original code from question
   distancesNetwork = np.array([])    
   for i in range(edges.shape[0]):
        distancesNetwork = np.append(distancesNetwork, \
        distance.euclidean(coords[edges[i, 0]], coords[edges[i, 1]]))
   return sum(distancesNetwork)

def vectorized_app1(coords, edges):
    return np.sqrt(((coords[edges[:, 0]] - coords[edges[:, 1]])**2).sum(1)).sum()

def vectorized_app2(coords, edges):
    s = coords[edges[:, 0]] - coords[edges[:, 1]]
    return np.sqrt(np.einsum('ij,ij->i',s,s)).sum()

-

In [114]: # Setup bigger inputs
     ...: coords = np.random.rand(100,3)
     ...: edges = np.random.randint(0,100,(10000,2))

# Verify results across all approaches
In [115]: networkLength(coords, edges)
Out[115]: 6607.8829431403547

In [116]: vectorized_app1(coords, edges)
Out[116]: 6607.8829431403337

In [117]: vectorized_app2(coords, edges)
Out[117]: 6607.8829431403337

In [118]: %timeit networkLength(coords, edges)
     ...: %timeit vectorized_app1(coords, edges)
     ...: %timeit vectorized_app2(coords, edges)
     ...: 
1 loops, best of 3: 519 ms per loop
1000 loops, best of 3: 822 µs per loop
1000 loops, best of 3: 668 µs per loop