How to tune hyperparameters for LSTM training.

I'm training a LSTM model for signal classificaiton. The training progress graph looks like the first pictiure at the very begininng and incrreases slowy. I think the model is over fitting or unddr fitting. I hope somone could resolve this issue.

[Badavath Purnesh Singh]

To let the algorithm train upto 95-99%, it is talking 3000 epochs. I have used 0.0001 as Learning rate and default mini batch value.

Answer 1

[David Willingham]

Hi,

I'd recommend using the Experiment Manager App for tuning / understanding various hyperparameter combinations.

Here is an example of how to use it for a LSTM problem.