Encoding finite numeric values with categorical or scalar?


One of my fetures is ID feature, this feature receive values between 0-28 (5 bits feature) , each value representing different person.
I have doubt, with which Encoder I need to work? with categorical encoder or with scalar encoder (with min/max limitations)?
In addition to the above , how much bits you are recommended to representing this feature in the SDR output (SDR n value) ?


I see a problem with your setup. If you have a ID in a row of streaming data, this usually means the data needs to be separated into groups by ID. Each ID should have it’s own model.

Unless the ID does not indicate ownership of the data in anyway, meaning it is truly just a feature or state of the data at that point in time, and does not indicate a classification.


The ID representing the person that take the action in the current step.
You can look on that as boarding game, that each person have turn for playing, but the turn is not round robin it’s mean player 1 can play 3 time while player 5 played only 1, in each time only one player can play.
The ID isn’t use for classification.


Ok got it. I suggest you use a CategoryEncoder. See: How to encode categorical data using CategoryEncoder


Thanks!! :