Training trouble with function with sharp edge

1

First I tried training a CNN with the function f(x)=x.
This worked fine.
Then my new function is f(x)=x, x<0.5
f(x)=0, x>0.5

This did not work, network could not figure it out.
Is that a weakness of neural network that it cannot model non-differentiable function, or what.

Thanks Matt

2

I think I have it working now. Maybe something I did wrong although I can’t figure out what it was.