Like many, I having trouble to make the loss decrease…
Everything seems fine in my code, I have no error and I double check with the former lab session on the Rubber Duck. Still, the loss doesn’t decrease and it drives me crasy…
Could someone please help me ?!
[code removed from moderator]
I am sorry but I removed your code because its not ethical to publish it here, now for this assignment which is a bit complex, the most probable issue is not selecting the right layers to further train on i.e. you are not setting the checkpoints correctly. I advise you first to go through the lab step by step again, in some cases it helps in getting it right, if you still not succeeding send me a copy on private I will try and have a look on it and hopefully find the mistake.
I had the same bug as OP described. Looking at the moderator’s comment, I found that I made a mistake naming checkpoint variables. This is within the last DIY code block of exercise 6.2. I wrote ‘_feature_predictor=’ instead of ‘_feature_extractor=’. It actually did not produce the expected output, but it slipped my visual inspection.
Hope this helps others.
[Update: model now trains as expected, with steady decrease in loss]