Here is a guide on how to deal with nan: http://deeplearning.net/software/theano/tutorial/nan_tutorial.html
Fred On Sat, Jul 9, 2016 at 4:34 PM, shashank gupta <[email protected] > wrote: > Yes, I tried batch GD and passed one example at a time. With that I am > getting 1-2 NaNs which I handled with try-catch block. But the cost is not > decreasing in that case. I am not sure if I have made some mistake while > writing the model or something else. > > > On Saturday, July 9, 2016 at 10:45:41 PM UTC+5:30, Dustin Ezell wrote: >> >> Very welcome. How large is the X you're passing in? try passing in just a >> few rows and see if you still get NANs. You may need to use batch gradient >> descent with smaller batches. >> >> On Friday, July 8, 2016 at 11:22:56 AM UTC-5, shashank gupta wrote: >>> >>> Thanks for replying. No I am running it on CPU (for validating the >>> model). >>> >>> On Friday, July 8, 2016 at 9:35:58 PM UTC+5:30, Dustin Ezell wrote: >>>> >>>> Are you running this on GPU? How big is X? I had a NAN problem from a >>>> cost function running batch gradient descent, that would only show up after >>>> several batches. Best I can recon my GPU is running out of memory since I >>>> can perform numerous iterations before running into issues. >>>> >>> -- > > --- > You received this message because you are subscribed to the Google Groups > "theano-users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
