Normalizing target and learner.show_results()

fastai
1

Hi,

I’m working on an Image Regression problem, my target values range from 20 - 2000 so I’ve been experimenting with using both y_range=(20, 2000) in the cnn_learner and also by normalizing the target values by subtracting the mean and dividing by the std in the DataBlock get_y function.

I’ve added a custom decodes to my loss function that reverses the normalization step, but when I call learn.show_results() I only see the predicted values have been decoded, so I end up with the target/prediction looking like this:

0.56754
(1028.60)

What is the correct way to work with normalized targets and then reversing them for interpretation?

Thanks in advance,

Kevin.

1 Like