self.optimizer = optimizer(self.model.parameters(), lr=self.learning_rate) # This is in class init function

# Training
self.model.train()
for iteration, batch in enumerate(train_set):

	# Next four lines of code prepare the inputs and targets
	landmark_batch = np.nan_to_num(batch[2].numpy())
	noisy_landmark_batch = add_noise_batch(landmark_batch)
	landmark_batch = np.array([np.concatenate((l[:,0],l[:,1])) for l in  landmark_batch  ])
	noisy_landmark_batch = np.array([np.concatenate((l[:,0],l[:,1])) for l in  noisy_landmark_batch  ])

	self.input.data.resize_(landmark_batch.shape).copy_(torch.from_numpy(noisy_landmark_batch) )
	self.target.data.resize_(landmark_batch.shape).copy_(torch.from_numpy(landmark_batch))


	outputs = self.model.forward(self.input)
	loss = self.loss_criterion(outputs,self.target)
	criterion_outputs += loss.data[0]
	loss.backward()
	self.optimizer.step()


#Testing
self.model.eval()
for iteration, batch in enumerate(test_set):

	# Next four lines of code prepare the inputs and targets
	landmark_batch = np.nan_to_num(batch[2].numpy())
	noisy_landmark_batch = add_noise_batch(landmark_batch)
	landmark_batch = np.array([np.concatenate((l[:,0],l[:,1])) for l in  landmark_batch  ])
	noisy_landmark_batch = np.array([np.concatenate((l[:,0],l[:,1])) for l in  noisy_landmark_batch  ])

	self.test_input.data.resize_(landmark_batch.shape).copy_(torch.from_numpy(noisy_landmark_batch) )
	self.target.data.resize_(landmark_batch.shape).copy_(torch.from_numpy(landmark_batch))


	outputs = self.model.forward(self.test_input)
	loss = self.loss_criterion(outputs,self.target)
	criterion_outputs += loss.data[0]