Hi all,
Is there any way to apply dropout to test data as well as training?I want to run multiple outputs on each test example so as to produce a range of results and calculate the mean and standard deviation with the aim of describing the degree of uncertainty?
Many thanks.
output has a few different signatures where you can say that you you want it to run in training mode. With that set, drop out will also be applied to test data.
Ahh yes I see those now. If I run output in training mode will this produce different results than if in test mode? Or is that just a silly question?
Many thanks again.
If you have anything like dropout in it, that typically gets disabled for inference, then yes you will get different outputs. Otherwise they should be equivalent.
Thanks. Yes, I want the only differing factor to be the application of dropout. So without it when calling output, running in train or test mode will produce the same result.