Topic suggestion for MLOps course

AI Discussions
1

Hi, I love your course on MLOps.

A problem I’m facing in different projects, is how one should split train-val-test data ?

For instance if you want to predict sales per store/item, should we split based on date only or should we split based on some stores/items and date or are there other best practices and patterns we can use.

This would be a great topic for your course because its a real problem for ml practitionners.

Best regards
Jean

2

train / val / test splits have nothing to do with features. It’s about dividing data based on the number of instances. Deep learning specialization goes into all the details on building a model.