I found numerous sigmoid-curve approximations for signum function. Some are really good, but a hard-sigmoid could be better, so does any approximations falling in that category exist?
And if a 'hard-sigmoid' approximation do exist, will that be better than it's sigmoid-counterparts?