I’ve been stuck on this problem for a few days now.
There’s 5 classes to predict. When my lr = 0.00001, the accuracy of class 0 is 99% and class 1 is 0%.
However if I remove class 0 and change the lr to 0.0001, class 1 goes up to 99% accuracy.
So I need a different lr for each class. I’ve never read about this problem, what should I do?
Is it possible to start with a high lr and once the accuracy of class 1 is above 98% stop considering it, decrease the lr to 0.000001 and only optimize for class 0?