平均列表 A 的所有元素，其索引在列表 B 中具有相同的值

Question

values_array = np.array([2,4,6])
coords_array = np.array([[127,130,130],[127,130,130],[128,131,132]])

Each element v in values_array has a "coordinate" c at the same position in coords_array.

I need a mapping from unique coordinate c to the average of all values which have that coordinate. For the example that would be

mapping[[127,130,130]] = np.mean([2,4])
mapping[[128,131,132]] = np.mean([6])

Without worrying about speed, I would do:

mapping = {}
for coordinate in np.unique(coords_array):
    indices = np.where(coords_array==coordinate)
    mapping[coordinate] = np.mean(values_array[indices])

I really need to do it without the loop in python if at all possible though.

Answer 1

We could use np.unique to tag each unique color and then bincount to get labeled average values -

In [145]: u,t = np.unique(coords_array, axis=0, return_inverse=True)

# Unique avg colors
In [146]: u
Out[146]: 
array([[127, 130, 130],
       [128, 131, 132]])

# Avg values    
In [147]: np.bincount(t, values_array)/np.bincount(t)
Out[147]: array([3., 6.])

Or additionally, use return_counts arg to get counts directly -

In [156]: u,t,c = np.unique(coords_array, axis=0, return_inverse=True, return_counts=True)

In [159]: np.bincount(t, values_array)/c
Out[159]: array([3., 6.])

So, if you need to assign these average values into a 3D array a , simply do -

a[tuple(u.T)] = avg # avg are average values from bincount output