KMedoids: add PAM algorithm

[rth]

Following discussion in #44 (comment) it could be useful to add partitioning around medoids (PAM) algorithm to KMedoids,

From wikipedia:

• Initialize: greedily select k of the n data points as the medoids to minimize the cost
• Associate each data point to the closest medoid.
• While the cost of the configuration decreases:
  • For each medoid m, and or each non-medoid data point o:
    • Consider the swap of m and o, and compute the cost change
    • If the cost change is the current best, remember this m and o combination
  • Perform the best swap of m_best and o_best, if it decreases the cost function. Otherwise, the algorithm terminates.

See "Clustering by Means of Medoids' Kaufman & Rousseeuw for more details. Also ESL book section 14.4.10 is worth reading.

[kno10]

Resolved by #73?

[rth]

Resolved by #73?

Indeed, thanks!