scikit-learn friendly Implementation of Neural Nets. Will also try to reproduce results in this area

Reproducing MNIST results in:

Hinton, Geoffrey E., et al. "Improving neural networks by preventing co-adaptation of feature detectors." arXiv preprint arXiv:1207.0580 (2012).

reproduced the results with the following two networks:

1. 784 X 800 X 800 X 10. 50% dropout on hidden units. No dropout on input units

• can be reproduced with python testMNIST_130.py

2. 784 X 800 X 800 X 10. 50% dropout on hidden units. 20% droupout on input units

• can be reproduced with python testMNIST_110.py