[BUG] - Loss calculation problem in train-loop

[jaggernaut007]

Add Link

https://pytorch.org/tutorials/beginner/basics/optimization_tutorial.html

Describe the bug

The loss calculation seems to be wrong. Should we divide the total loss by the number of values of X rather than the number of batches?

The optimiser also seems wrong.

Describe your environment

Running on Google Colab

cc @subramen @albanD

[jmarintur]

Hi @jaggernaut007, are you referring to this line? If that's the case, the loss within the testing loop gives you a single scalar value per batch, and that's what you are accumulating in test_loss. That's why you need to divide after the loop by the number of batches, num_batches.

[jmarintur]

@svekars unless I am wrong, I think we can close this issue.