Is this training graph looks okay?

Hi there, I am training my PyTorch (1D CNN )model for 50 epochs on an audio dataset. the first graph is without using any validation loss scheduler while for the second one I have used a loss scheduler having patience equal to 9. The second graph is quite improved as per my observation, But still, I want some good comments on this, what else I can do to improve my classification accuracy here.
Plz, suggest.




It is not necessary that you will always get good test accuracy.

Although there are many things you can try

  1. Better scaling of the dataset (Normalization)
  2. Drop out or batch normalization (I guess you should try this, like in your cases, overfitting)
  3. By varying scheduler learning rate.

Thanks for reply @Kapil_Rana sir I have changed the first graph though , now its correct.
Also I am already using step number 2.
and graph two is the result of step 3, So I that’s shown the results is improved.

Training is okay in the graph, but validation accuracy is not improving, it is the case of overfitting.

You can try a smaller model, a simpler one.
start with a small model and gradually make it bigger. and check on results.

1 Like