Feature scaling: scaling one mean scaling all?

If I scale one feature, does that mean that I also have to scale all of the other features?

You only need to normalize features that have a wide range of values.

Technically you do not have to normalize all of the features - but it is much more convenient if you use a function that normalizes all features, so you don’t have to create a custom method that allows you to specify which features to normalize.

Normalizing features that don’t really need it doesn’t cause any harm, it just consumes a bit of computer resources.

The one exception is that normalzing a feature that is a constant is going to cause a divide-by-zero error, because the std dev of a constant is zero, and you normally scale by 1/sigma.