Update dcgan_faces_tutorial.py (#816)

* Update dcgan_faces_tutorial.py

Previous comment might have been slightly misleading. A bit easier to understand, if made explicit, that gradients of errD_real and errD_fake w.r.t. parameters of netD get added up/accumulated because of the successive backward calls without a zero_grad() in between.

* Update dcgan_faces_tutorial.py

Co-authored-by: Brian Johnson <brianjo@fb.com>
Co-authored-by: holly1238 <77758406+holly1238@users.noreply.github.com>

Author: 3 people

beginner_source/dcgan_faces_tutorial.py

@@ -610,10 +610,10 @@ def forward(self, input):
         output = netD(fake.detach()).view(-1)
         # Calculate D's loss on the all-fake batch
         errD_fake = criterion(output, label)
-        # Calculate the gradients for this batch
+        # Calculate the gradients for this batch, accumulated (summed) with previous gradients
         errD_fake.backward()
         D_G_z1 = output.mean().item()
-        # Add the gradients from the all-real and all-fake batches
+        # Compute error of D as sum over the fake and the real batches
         errD = errD_real + errD_fake
         # Update D
         optimizerD.step()