Dear Ayman,
Thank you for your interest in our work.
In the parse we normalize traffic and delay according to the mean and standard deviation of such variables in the training dataset. This is a satandard procedure in Machine Learning. Hence, we then need to denomalize the output values of RouteNet to get the real delay predictions:
predictions = 0.54*preds + 0.37
If you want to predict jitter, you would need to train the model with the jitter labels. To accelerate the training, you could start with a model already trained on delay labels, as jitter is somehow correlated to delay.
Regards,
José
Hello,
I am a student, and I am trying to recreate the paper, to see the results I have tried to use the notebook that you provided, there is something that I don't understand,
I see that you are normalizing the delay and traffic in the parse, and after you get the predictions you are normalizing them using:
predctions = 0.54*preds + 0.37
why is that?
and if I want to predict the jitter for example do I need to train another network with the labels being jitter?
thank you
_______________________________________________
Kdn-users mailing list
Kdn-users@knowledgedefinednetworking.org
https://mail.n3cat.upc.edu/cgi-bin/mailman/listinfo/kdn-users