Implement Decay for Training Data

First idea is to remove weight periodically (or after n amount of new data) and then remove nodes if no weight is left.