Support Out of Memory Solutions

[jgilme1]

Many of the operations involved in training the Multir Model rely on in memory solutions that will not scale to larger feature sets. and larger amounts of training data.

[jgilme1]

A simple NoSQL embedded Java DB should work, maybe OrientDB