We already know that non-real-valued fields of a data set to be modeled by neural network or some other machine learning approach must be transformed into real value. Some categorical or nominal type field may be one-hot-encoded and some natural language token may be embedded into vector.
What about for already real-valued data field? If I want to build a regression model using
I don't have to normalize the data set further because they are all real-valued? The reason for posting this is that when there is a big difference in magnitude of each dimension, it feels like I have to re-scale them to have the same min, max value, for example, 0.0~1.0.
Gist of my question : There is no need for transformation for already real-valued data field to map them into vector space?