I have a list of possible transforms, and I've read some confusing/contradictory stuff about the preferred order in which these operations are performed. Maybe 1) the order is sometimes amorphous, and 2) some of the transforms are mutually exclusive? Clarity in those ways would be useful, too.
So, in my very rough proposed order as I understand it (not meant to be a comprehensive list, and if I am missing something, please let me know):
1) Standardize dates 2) Categorical variables to multiple binary (dummy) 3) Discretize any quantitative variables you want 4) Missing Data Imputation? Should this go above/below? 5) Scale and center 6) PCA/similar dimensionality reduction? 7) Forward/backwards selection? 8) Regularizers? Order for 7) versus 8)? Does 7) imply 8)? 9) Sub-sampling rows (should this actually be at/near the top (first)? could it be either first or last?)
Bonus points if you can point me to a great conceptual explanation of how common econometric tests for robustness and autoregressive functional form fit into the above timeline of operations. A simple list that gives me a timeline for both ml-style transforms and econometric tests/transforms, integrated in one place, is useful too, but I'd prefer a great explanation/link.
I apologize if I'm asking for something inappropriate.