Hi,

Thank you for your nice codes and it works quite well on MNIST reconstruction!

However, I do notice a very interesting phenomena: Although the training seems to be successful and the visualization of reconstruction looks good, the loss is increasing during training. But it seems that the optimizer in pytorch should always lower the training loss. I am very curious about the reason, and would you please explain why?

I also print the L1loss between the original images and the reconstructed images, and the L1loss is decreasing as expected.

Best regards,
Wang

Hello~

Thank you very much for sharing the code, which really helps me a lot.

But I find some problem when I use the code in

RBM.ipynb

def free_energy(self,v):
	vbias_term = v.mv(self.v_bias)
	wx_b = F.linear(v,self.W,self.h_bias)
	hidden_term = wx_b.exp().add(1).log().sum(1)
	return (-hidden_term - vbias_term).mean()

I found that there may be numerical instability in free_energy. In hidden_term = wx_b.exp().add(1).log().sum(1), if wx_b contains elements that greater than or equal to 88, the result of wx_b.exp() is inf. And this may cause the result of loss to become NAN.

It could be fixed by making the following change:

hidden_term = wx_b.exp().clamp(max=87).add(1).log().sum(1)

Hope to hear from you ~

Thanks in advance! : )