Questions on Training count and Loss Rate
sreenathpv1989 opened this issue · 0 comments
sreenathpv1989 commented
Hi
I trained model with 500000 times as mentioned in TRAINING file the final loss data is 1.5736 as below
Epoch 120/120
225/225 [==============================] - 20s 91ms/step - loss: 1.5736 - msse: 0.0511 - val_loss: 2.2502 - val_msse: 0.0684
But with test data, my model is not able to remove the noises clearly. But when I used default rnn_data.c available in this project the results are quite good.
- It seems model not learned my noise well, What are the expected value of loss , msse for a good trained model ?
- Since the count 500000 not seems enough, How many times we need to mix noise with clean signal for a good trained model?
- what are the values used & obtained for count and loss for generating the rnn_data.c in this package