deep-rules
deep-rules copied to clipboard
Normalize, normalize, normalize.
I think this would be a rule to talk about feature scaling, batch normalization, etc.
I think this need a corollary:
Normalize, normalize, normalize (...but don't normalize your data prematurely).
I think every data scientist goes through a phase were they just throw some sort of normalization / whitening / SVD at the data before they even look at it in all its raw glory. I certainly did, and I have seen pipelines where basically every second step was some sort of normalization step. Absolute scale is sometimes a feature, too!