Sigmoid Function to Rescale Feature

Good day,
I have just known the Sigmoid Function, & i was wondering if it can be used to rescale the feature instead of Z-Score and other methods.

I’m asking that because i find it very interesting function, unlike other rescaling methods, I don’t have to know anything about the data like max, mean or std.
I’ve tested it but it didn’t work as expected, the GDA didn’t converge no matter what alpha and iterations i use.

Hi @ahmedelmasry1985, the rescaling methods covered in the course are all linear transformations which is safe to use. Sigmoid is a non-linear transformation and it won’t preserve the linear relationship between your feature and your label. Let’s say you have this dataset of 1 feature and 1 label, one followed by another:

3, 6
5, 10
8, 16

Very obviously it is a linear relationship between the feature and the label y = 2x. However, if you transform your feature with sigmoid, the transformed feature values will become essentially 1

1, 6
1, 10
1, 16