Picking Vectors from the Clusters

I need to pick 360 vectors using the 2500 Clusters.
I'll pick the "best" vector from the top 360 clusters.
To pick the best vector from a cluster, I will:
- calculate the centroid of the cluster
- pick the unlabeled vector that is most similar to this centroid
- in case of a draw, I'll pick one arbitrarily.

The clusters vary in size from a low of 51 to a high of about 1200. On an average, there are approximately 400 vectors per cluster. So the 360 clusters comprise of approximately 144,000 vectors. Picking representatives from each of these clusters should help the transductive learner.