How to effectively test and debug machine learning models, from ML engineer@Apple to startup founder - Gabriel Bayomi - the data scientist show #055
The Data Scientist Show - Daliana Liu
00:00
Do You Need to Update Your Validation Data Set?
So I agree that you should update your validation data set to make sure you represent the real world. But do worry when you add new examples based on your understanding of the real world, do you adding bias to the validation set? That could definitely happen. So always be aware that your validation set should not be static. Things important, but it should also think about, you know, issues of bias and other things that can come in the process. Yeah. And previously we talked about like tagging the data for a segment. So later on, it's easier for us to do our analysis. For example, do you create a new feature and then just add basically your own labels to it?
Transcript
Play full episode
