When your dataset is ready for Model Experiments, it’s recommended to generate train & test metadata beforehands.
- Go to the Model Experiments view
- Click on the split button at the top right
- Results will be in ‘Corpus’ metadata accessible in:
- the Documents view with the filter panel for Text classification project
- the Segments view with the filter panel for Entity detection project
- Note that when you start a new model experiment, the split will be automatically updated if you have added new annotations in the dataset