16. Splitting data in train/val/test splits
When we evaluate and train our models, we need to do so on separate training, validation, and test data sets. In this video we describe the best ways to split this data up and some things to keep in mind as we try to avoid data leakage and poor model generalizability.
Check out the whole materials informatics series at • Materials Informatics with workbooks and course notes available at https://github.com/sp8rks/MaterialsIn...
0:00 data splitting to prevent p-hacking
4:50 train/val/test data sets
7:00 validation sets for model tuning
9:00 leave one cluster out cross validation