the implementation of utils.py/Sampler refers to https://github.com/phlippe/uvadlc_notebooks/blob/master/docs/tutorial_notebooks/tutorial8/Deep_Energy_Models.ipynb
usage: python train.py
some sampling results after training 100 epochs:
loss trend:
energy of real samples are slightly lower than fake samples
contrastive divergence keeps fluctuating around 0 while regularization loss converges to 0 soon
