After quite a marathon these last days, i’ve just reached the 60%. This course was a lot of fun and I can only continue after the deadline towards full completion, otherwise what’s the point
Videos and notebooks were made with care by the whole team and this is greatly appreciated.
My feedback “à chaud”:
-
I remember being surprised to read about hyperparameter tuning in Module 3 while Module 2 exercise and quizz already significantly dealt with them. This kind of surprise: “Oh, i’ve learned that by myself yesterday, for the quizz from before”
It went fine for me though. -
Overall, the hardest part, to me, often was getting the results out. Should it be this way? Model, check, pipeline, preprocessing, check, cross-validation, check… Now getting the features and weights out of the pipeline results object -a very common analysis protocol-, requires what feels like way too much error-prone wet work through dataframe comprehension and reconstruction. I have tried to make a function for myself but this all depended too much on the pipeline composition, apparently, for a simple def to handle it.
Do you plan on easing estimator extraction in the future? Otherwise i’ll find a way to do it.