Open rohithkodali opened 6 years ago
In order to improve the training speed, my model is very simple, which is enough for a simple dataset, but may be under-fitting for a larger dataset. And I only use logistic regression to implement the attention mechanism. So if you use more data for training, you can try adding more hidden layers and implementing the attention mechanism in a more complicated way.
HI,
When i have added more data for training (around 20 hours) the val_acc is too low. it is around 0.01943. Am i missing something while creating the model instance i mean do i need to add any hidden layers or to change any activation functions.