I am trying to calculate the accuracy and loss for the test dataset. Since we dont have labels for our test dataset, what kind of loss does the “get_preds()” function return for the test dataset?
Thanks!
The test set is pseudolabels, eg everything is 1. So the ‘loss’ isn’t really a thing you just care about the predictions.
Do you have labels beforehand? If so there’s a different way to go about it.
Thanks for the reply. I have the labels beforehand too. How do I go about that?
I have exported my learner as well. I need to add the test dataset with labels. Can you please help?
Sure! Here’s a notebook link however if you need more guidance when I’m at a computer I can show code. Is it tabular or image.
I am using image dataset with classification labels. I’ll take a look and let you know!
Thanks!
Sounds good 
the above method works for images too.
1 Like
I also had another question. When I predict for a test img with labels and calculate the loss, I want to backprop this loss to the input image and calculate the total error at the input layer. What kind of parameters can I pass in “learn.backward(item)”? Is this the right way to go about this?
That one I don’t quite know the answer to my apologies 
hopefully someone can chime in with that!
No worries