Data Science and Machine Learning (QLS-DSM) Lecture 2

okay i'm recording yes please can you repeat yeah the hype hypothesis class here also means model class yeah model class and hypothesis class is the same [Music] model class or hypothesis classes so the question now is um how to obtain these parameters okay and so we def we need to define a notion of goodness okay to select these parameters uh and to select the function in our hypothesis class so the third ingredient it's a notion of goodness okay and this is formalized thanks to a cost function or just cost or energy function okay you can there is no theorem here there is no rule this is your choice okay and what is important is that this splitting into training and test data okay the splitting is done randomly okay so you this would be a absolute mistake to try to select by hand or according to some external procedure which point you think are good points in order to train your models and which one you should use for testing you should never do that you have to select them totally randomly otherwise you are introducing bias okay it means that you are already assuming something about the underlying system that is not captured automatically