Current MultiLayerPerceptron has only implemented the basic features of a Multi Layer Perceptron.
There are still several features need to be implemented.
In the next step the following features should be added:
1) add more cost functions such as cross entropy.
2) add momentum and regularization.
3) make the training method in MLP be public to allow user to use MLP in a standalone algorithm.
4) add more test cases on other applications, at least one with regression and one with classification.