> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/sdgm/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/sdgm/_mcp/server.

# Baselines

## **Baseline Evaluation Metrics**

On your **training job's evaluation page**, Kumo provides **comparisons of your model versus heuristic baselines** for **temporal binary classification, link prediction, and regression tasks**. These baselines allow you to compare the results of your **Kumo model** against **simple heuristic-based approaches**.

Baselines are computed on a **per-entity level**, meaning:

* **Regression tasks** use **past labels as forecasts**.
* **Link prediction tasks** recommend **past user behavior** as future predictions.

  ![Screenshot2025 06 27at10 53 33AM Pn](/sdgm/_files/nvidia-sdgm.docs.buildwithfern.com/327cf22b3ea6517dbfa8b9d7d0a29f318dcffb4a95af8e8b2b0fbf05bffad379/images/Screenshot2025-06-27at10.53.33AM.png)

Baselines are not supported when `timeframe_step` is `None`.

***

## **Baselines Per Task Type**

The following table outlines **current baselines** supported by Kumo, organized by **prediction task type**:

| Task Type             | Temporal Problem                                                                                                                                                                                                                      |
| --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Link Prediction       | Per Entity Frequency – Predicting each entity with the most popular ranked list of target values from all previous splits.                                                                                                            |
| Regression            | Per Entity Mean (Personalized Average) – Predicting each entity with the average of all past target values from previous splits. Per Entity Last Label – Predicting each entity using the **most recent past value** as the forecast. |
| Binary Classification | Per Entity Mean – Predicting each entity with the average of all target values from previous splits. Per Entity Last Label – Predicting each entity with the previous time frame’s label                                              |

### **Why Use Baselines?**

Baseline models provide a **reference point** to determine how much your Kumo-trained model **improves over simple heuristics**. A well-performing model should **outperform these baselines** significantly.