Measuring the distance between two distributions of vectors

I have 2 buckets.
Each bucket holds documents.
Each document has a vector field.
I want to measure the distance between them.
To measure the distance, I need to take the mean vector out of each bucket , and calculate the delta between them.

What the best way to save the vectors ( knn_vector ? spread the vector to discrete field x0 , x1 , x2 ?) and perform the calculation of the mean vector? (Painless script that iterates the vector, save in the state the sum of each dimension and calculate the average in the reduce_script ?)

Sorry for getting late here. For these kind of operations, it would be better to do computation at client side. Painless scripting is one way to do, but might be complex.