I’m trying out the week 4 ungraded lab 2 with a GPU runtime, and I’m getting very bad results when I train: every image gets labeled as a cat. What did you change to get better results?
I want to play around more to try to fix it, but Colab kicks me off for using too much compute.
Are you using your own images for inference?
No, I ran the notebook as it was provided in the course, without a single change, and it didn’t work.
I tried to make some changes afterwards to get it to work, but I ran out of compute.
Never mind, I reran yet again, and it now classified some dogs, though it’s still quite inaccurate.
1 Like
honestly I felt when I ran that assignment while my learning phase, the reason for inaccuracy would less number of images or data to train on.
The issue you first mention encountered about colab not working is because this course specialisation assignment can run the model for 2 or 3 times max after that for free tier colab, I had the same issue.
Did you try using the tpu?
I did try the TPU, but that ran out of compute even faster 
tpu usuage has 24-hour cycle. if you have used up, used next day or use paid colab