stopping rules for model training

[sonalgoyal]

We need to provide the right signals to the user in terms of how the model is converging and how it is performing on the training. right now it is pretty difficult to assess when to stop.