I’m running the Transfer Learning with MobileNetV2 assignment. All of the inline tests pass and the grader gave me a 100 when I submitted the assignment but the results of model2.fit look wrong to me.
After running the fine tuning code, the accuracy doesn’t change and the loss continues to get worse.
Any thoughts on what I might be doing wrong?
I agree that something looks wrong there. For starters, “loss” values should never be negative. But we are using TF logic for the loss functions. Are you sure you are running this on the Course website as opposed to your own local environment?
Yes, this is running on the Course website.
I don’t know what the problem is, but it seems worth investigating. Check your Direct messages.
In case anyone else runs into this, it was an environment problem.
The problem was caused by the presence of an extra subdirectory, .ipynb_checkpoints, in the dataset subdirectory. When I executed the image_dataset_from_directory function it showed 3 classes, not 2. I created a terminal window then removed the subdirectory using the command, “rm -rf .ipynb_checkpoints” from within the directory subdirectory. This allowed my code to run correctly. Thanks to Paul for his help.
