Currently, I'm working on my very first classification project. If you want to know what dataset I'm working with, think "playing stairway to heaven in your local guitar store", and it will probably come to you. But whatever, I'm a newb and I have a question about best practices:
During feature extraction, do you:
- extract one feature at a time, test the model performance and then choose to keep it or discard it?
- extract a bunch of features at once and just throw them all at the model because doing one at a time takes too long?
Currently, I've been doing one feature at a time, but my notebook is getting suuuper long (lots of failed strategies and only one feature has added value). I am running into the same issues expressed in this question.
His/her question is different than mine, but similar in nature.