Hi, are we supposed to achieve 99% training acc and 95% val acc with limited parameters which is 30000? It worked obviously using higher values of filters for conv and I’m not sure if we’re allowed to do that. I tried using adam and RMSprop for optimizer as well as modifying dropout values but none of them worked.
Hi @penguin12
If you want higher accuracy, you can increase the model capacity, tune hyperparameters, or train for more epochs. Generally, if you follow the same approach as in previous ungraded labs, you’ll receive full marks.
Hope it helps! Feel free to ask if you need further assistance.