Hierarchical classification
Walk a deep taxonomy to the right leaf with parallel Choice questions.
Classification- Result
- Leaf labels from parallel beam search over Choice probabilities
Taxonomies, org charts, and code trees are Choice questions stacked in a path.
Greedy search takes the best child at each node. Beam search asks several paths in one Jev call and keeps the best K.
The cookbook reports this as the way to classify documents that live in deep, named hierarchies.
Pipeline
- Start at the root of the hierarchy
- Ask Jev to score the next-level options
- Keep the best K paths by geometric-mean probability
- Stop at a leaf and return that label
Builder: TypeSafe cookbook. Stack: jev.