I am trying to implement an ANN, which has n different variables taken in input. I want them to be scaled in a range between 0 and 1. My question is: should I scale each column of input individually or together?
Thank you for your help.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
0
The answer depends on your data.
In case if type of the data is different in the different columns then it's better to scale each column separately. For instance, let's say you have to input columns the first one is age of a person (from range 18 - 80 years) and the second one is a house price (from range 100,000$ - 1,000,000$). If you scale them together then your years will be vanished on a scale of house prices, because values in the year column are much smaller.
In case if all values are the same then you can scale them together using the smallest and largest values from overall dataset. For instance, you have a dataset of images 8x8 pixels. At first yo can convert it to a vector with 64 values. Each value can be between 0-255 (white-black image), but it's not necessary that in each column you will find some pixel that equal to 0 or 255 (min and max values). So in this case you can scale all values together.
