Minor documentation fixes

[theodorju]

Hi, all

I've been reading the documentation and had the following findings in the code for [this] (https://automl.github.io/HpBandSter/build/html/auto_examples/example_5_pytorch_worker.html) example. More specifically:

• The loss is set equal to zero at the start of the for-epochs loop, but later it is re-calculated from the start The loss=0 statement right after the for-epochs loop could be removed.
• Criterion is set equal to criterion = torch.nn.CrossEntropyLoss() but then it is never used.
• Function evaluate_accuracy has an import pdb statement commented out. Not sure if this is intended.
• Function compute has a # device = torch.device('cpu') statement commented out. Not sure if this is intended.

Additionally in this example the line inc_run.info['test accuracy'] should be replaced with inc_run.info['test_accuracy'] (underscore needs to be added).

I would like to work on fixing these minor changes as soon as I have some time.