keeping the validation loss stable was easy enough and I have successfully completed the assignment. Just out of curiosity, I have been trying to score a good validation accuracy score too. However no matter what I try, it doesn’t seem to go higher than .73~.77. I guess this isn’t a “I have an issue” post, but I was just wondering has anyone able to top that, and if so then how?
There are several ideas you can try: 1. Change the structure of the model (including the number of neurons and network layers). 2. Reduce the learning rate. 3. Change the optimizer. 4. Increase the number of epochs.
2 Likes
The ideas suggested by @SteveSun should help you
Is it also common to combine LSTM, Conv1D, GlobalMaxPooling1D and Dropout in building a model architecture?
I thought we EITHER choose LSTM or convolutional layers (as a simpler approach). Therefore, combining such layers would be quite common!
Is that idea quite legit?
Cheers,