This part (1:02:45) is not clear to me: "if you need a final model (for production or for inspection) fit the model using the 'inner loop' on the entire dataset" Could someone shed some light? How do I get the final model?
I think what he ment is that after you know the best lambda (or other hiperparameter) you split the whole data on trening and validation with the idea to generate the final beta coefficents and lambda. You don't need test set at the end.
Absolute goldmine
This part (1:02:45) is not clear to me:
"if you need a final model (for production or for inspection) fit the model using the 'inner loop' on the entire dataset"
Could someone shed some light? How do I get the final model?
I think what he ment is that after you know the best lambda (or other hiperparameter) you split the whole data on trening and validation with the idea to generate the final beta coefficents and lambda. You don't need test set at the end.
Dmitry thank a lot for these lectures!
Any particular resource where I could read more detailed about this?
It seems to me that this course is highly influenced by ISLR. You might wanna take a look at that.
Which book do you guys follow?
Are there slides for these lectures?