I’m a bit confused when deciding to standardize the features.
I already understand the purpose of this technique, but I get in trouble when deciding which features to be standardized.
Should be all features standardized or just the one that reflects the bigger scale discrepancy?
The example in lecture “To Standardize or to not Standardize” got me a bit confused.
It really depends on the model.
If you are looking for interpretability – you don’t want to standardize any features.
If you are looking for predictive power, you want to standardize all of them.
You are implying that there could be a mix. Yes, this is usually done when you’ve got dummy variables. You’d standardize all features, but the dummies.
The 365 Team
great! thank you