Question

I Have a N x D dimensional features, which I need to rank according to their distance to a 1 x D dimensional vector. Any fast way to implement that in python without recursively apply argmin?

Thanks!

Answer 1

Something really simple is Squared Euclidean Distance, and it’s implementation would be like:

In []: F= randn(5, 3)
In []: t= randn(1, 3)
In []: ((F- t)** 2).sum(1)
Out[]: array([  8.80512,   4.61693,   2.6002,   3.3293,  12.41800])

Where F are the features and t the target vector. Thus the ranking would be:

In []: ((F- t)** 2).sum(1).argsort()
Out[]: array([2, 3, 1, 0, 4])

However if you are able to describe more on your case, there might exist more suitable measures, like Mahalanobis distance.