DCGan MNIST Dataset Attempt.

This is an attempt to run a Deep Convolutional Generative Adverserial Network using Mnsit Data as basis.

The procedures are as follows:
A) Updating the Discriminator
1) generate fake noisy and random image, feed it into generator and make a fake image.
2) ask the discriminator to predict and get the criterion loss for fake images.
3) ask the discriminator again to predict the real images,
4) get the criterion loss for real images.?
5) get the average of the criterion loss for real and fake images.
6) Update the discriminator weights through back propagation and the optimizer through step.
B) Updating the Generator
1) regenerate fake noisy image, pass it in generator to get fake images.
2) ask the discriminator to judge whether its real or fake.
3) get the criterion loss based on fake images.
4) re-discriminate real image
5) get criterion loss based on real images.
6) average the criterion loss,
7) backpropagate the generator's weights
8) update generator's optimizer through step.
Repeat based on number of epoch set.