Adam Optimizer Tanking Model

I have been trying to implement my own cat recognizer model based off the coursework. I took the cat recognizer 2.0 and have been trying to improve with with techniques learned in the second course. However, whenever adding the Adam Optimizer to the model, the model performs dramatically worse than before. I am unsure why this is happening … my suspicion is that there is some evil little bug. The cost does not drop at all and the model isn’t learning.

My model with L2 regularization performs as following:
Train Accurary: 99.99
Test Accuracy: 84.00

Adding Adam Optimizer performs as following:
Train Accurary: 65.55
Test Accuracy: 34.00

I have debugged for quite some time now, but I cannot see anything wrong. Any insights into this problem, regarding code or the optimizer itself would be appreciated.

Here is my Github which is intended only for my personal use for this course and not open source to the world: https://github.com/jdeda/CatRecognizers.git

Hi Jldeda,

Kindly share the traceback/ error log you are getting.

Thanks.