Normalize Statistic models to use a common convergence controlling mechanism

[GoogleCodeExporter]

The learning of Logistic Regression models is done in the same fashion as 
AForge.NET's Neural Networks: a Run method has to be called to run iterations 
(or batch of iterations) manually, until the user decides convergence has been 
reached.

The Cox's models, the Hidden Markov Models and Hidden Conditional Random 
Fields, on the other hand, control the iteration themselves. You specify the 
number of iterations and tolerance threshold and call them once.

It is needed to decide which is most appropriated and enforce this as a common 
standard for all learning methods.

Original issue reported on code.google.com by cesarso...@gmail.com on 7 Nov 2012 at 11:07