Hi, My train_step() function is returning too high total cost (AssertionError: Unexpected cost for epoch 0: 2078114.625 != 25629.055). I am computing a_G by passing ‘generated_image’ to ’ vgg_model_outputs. And Then using a_G, and precomputed a_S, and a_C to compute the costs as suggested. All the cost functions passed the tests. Any clue what might be wrong?
One problem is that this notebook is “stateful”: you can’t run train_step several times and get consistent answers. The first thing to try is:
- Kernel → Restart and Clear Output
- Cell → Run All
Then check your outputs from the test that runs train_step.
If it’s still wrong, then we can go to the next step.
Well, maybe that theory won’t help. I tried a few experiments with running train_step several times and I get wrong values, but not “crazy high” values like the one you show. So I think there is probably something more wrong in your code.
The issue was similar to this: Week 4 Assignment 2 Exercise 6 Issue
And, it worked.
I’m glad to hear you were able to find the solution. Sorry that I didn’t think to suggest the integer arithmetic issues. Normally those throw negative values. Here is a thread which goes into more detail on that.
1 Like