Hi Friends,
In the week 1 programming assignment, i noticed 2 checks around the standard deviation of the output of the neural net (see below). How can we know that a neural net that is initialized will have a standard deviation between 0.55 and 0.65?
assert test_output.std() > 0.55
assert test_output.std() < 0.65
It’s not any neural net that will have that property. You have to carefully study how they define everything in those test cells. The tests are carefully designed to check that your code is behaving as expected, but sometimes they are “too clever by half”. 
Note that there is default behavior in pytorch for how weights are initialized and you can override that as well. So you have to check what is done in each case.
Thankyou!