Clustering Datasets in H2O Label Genie
- Predictive AI
- H2O Cascade
This lesson from the Label Genie Starter Course, part 8, shows how to use the clustering feature to explore a dataset before annotating it. The presenter loads a Disneyland review dataset and a fish image dataset, then creates clustering tasks with Gaussian mixture or K means models to group similar text or images. He shows how to inspect clusters in a 2D or 3D map view and step through examples with keyboard shortcuts, demonstrating how clustering reveals patterns like sentiment groupings in reviews or species groupings in images. The lesson frames clustering as a fast way to get a feel for a dataset before annotation.
Key takeaways
- How to create a clustering task in H2O Label Genie by choosing a number of clusters, an embedding model, and a clustering method
- Why clustering helps you understand a large text or image dataset faster than reading through every row manually
- How to use the K and J keyboard shortcuts to jump between clusters while reviewing examples in the gallery view
- What the map tab shows: a 2D or 3D visualization of where each cluster sits relative to the others
- How clustering a labeled fish image dataset by species count validated that similar images grouped together correctly
Start owning your AI