Closed sagarhukkire closed 7 years ago
Hello,
you can do less iterations. that’s for sure. one thing you should try is to do for example 30000 iteration and put step size 10000. In this way you will decrease the learning rate every 10000 iterations, by default by one order of magnitude.
regards,
Fausto
On 30 Mar 2017, at 01:18, sagarax009 notifications@github.com wrote:
Hi
In VNet there are 100000 iteration and stepsize is 20000. In my cases I saw I am getting convergence of loss withing 15000 something so can I train for such limited number of iterations? And if yes how the iterations and stepsize values are related is any math behind it?
Thanks Sagar
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/faustomilletari/VNet/issues/23, or mute the thread https://github.com/notifications/unsubscribe-auth/AMtsvg4NcPhK_pmar8Y57Lhu_JbJ1Cvqks5rqubVgaJpZM4MtrSN.
Thanks @faustomilletari
Hi
In VNet there are 100000 iteration and stepsize is 20000. In my cases I saw I am getting convergence of loss withing 15000 something so can I train for such limited number of iterations? And if yes how the iterations and stepsize values are related is any math behind it?
Thanks Sagar