Add a central capabilities overview for the native cuML API so users can see what cuML provides and how completely each estimator or function is supported.
For each public estimator or function, document:
- Supported input types, including dense and sparse inputs.
- Single-GPU, multi-GPU (MGPU), and multi-node multi-GPU (MNMGPU) capabilities and their limitations.
- Links to the corresponding API reference and detailed documentation.
Where a cuML API corresponds to scikit-learn, umap-learn, or hdbscan, intentional deviations should be documented both in the capabilities overview and alongside the affected estimator, method, or function. These comparisons should clarify the cuML contract without becoming the primary focus of the documentation.
Acceptance criteria
- The overview covers all public native cuML estimators and functions.
- Sparse-input support and other important capability boundaries are visible at a glance.
- MGPU and MNMGPU support and limitations are clearly identified.
- Detailed API documentation describes relevant limitations.
- Intentional deviations from corresponding upstream APIs are documented at the relevant API surface and linked from the overview where applicable.
- The overview is organized around native cuML capabilities, not upstream parity.
- The documentation defines how this information remains current as support changes.
Add a central capabilities overview for the native cuML API so users can see what cuML provides and how completely each estimator or function is supported.
For each public estimator or function, document:
Where a cuML API corresponds to scikit-learn, umap-learn, or hdbscan, intentional deviations should be documented both in the capabilities overview and alongside the affected estimator, method, or function. These comparisons should clarify the cuML contract without becoming the primary focus of the documentation.
Acceptance criteria