ML pipeline order

This course teaches that you always build a base model, then bias/variance analysis and error analysis will tell you the direction to go. And you use hyperparameter tuning with cross validation to make adjustments. I’m confused on the specific order of these steps when put into practice. For example if I just built a base model, do I tune hyperparameters first, then do bias/variance analysis? Or do I do bias/variance analysis, then tune hyperparameters? Initially, I may want to tune all hyperparameters, but if during bias/variance analysis I discover that I want to modify a few specific hyperparameters, do I go back and only retune those specific ones? Just trying to figure out how these are put into practice.

Hi @Max_Rivera,

I believe Andrew talks about “best practices”, so these are not “written on stone” kind of steps. As a ML practitioner, you can follow your intuition and come up with your own version of analysis.

Best,
Mubsi

1 Like