standard_parameters will fail when alpha==0

[pberkes]

I forgot to test for the corner case / not unlikely case when alpha is exactly 0.0 .
The standard_parameters method will fail for those cases, because the rescaling will give inifinities and nans.

other stuff would have failed also before #169, for example in the Logistic sigmoid we had
C = 2 * np.log(1 / self.alpha - 1) / width
which is going to cause a division by zero.