I’m currently on Chapter 7 in the book where we are learning about techniques used in State of the art Image Models.
Jeremy mentions about the importance of prototyping on a small dataset if your dataset is large (and thus, why he created Imagenette).
Here’s my question, how exactly do we make a subset of a dataset? And what are the rules to make sure its representative of the rest, such that the trained model will adapt well to the full dataset eventually?