> 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.

# Choose Table Columns

> Select clean, relevant, and sufficiently complete columns for AutoML

> **Prerelease documentation**
>
> This Kumo Studio information is prerelease and is not yet supported. Stay tuned for the Docker and Kubernetes releases.

Select columns that meet all three of the following criteria.

* **Clean:** Remove the following types of unreliable data:
  * Fake or synthetic data
  * Predictions from other ML models
  * Columns whose definition has changed over time, especially where the same attribute ID may point to different things at different times
  * Data that is otherwise unreliable or frequently inaccurate

* **Relevant and Mutually Exclusive:** The larger the graph size (the sum across all tables in a graph), the higher the compute cost.
  Remove columns that provide duplicate or similar information, irrelevant information, or other extraneous data.

* **Complete:** Cover the full history across the timeframe in question. For example, include the complete record of purchases or interactions rather than a user's first or last purchase, or a subscriber's most recent interaction.
  If this results in an oversized dataset, provide Kumo with a compressed version that indicates changes in aggregate metrics over time (for example, per day, week, or month).

> **Warning**
>
> Using the wrong or unnecessary columns can lead to both degraded model performance and increased training costs.

See [Data Preparation Best Practices](/kumo-engine/1.0.0/kumo-studio-guide-coming-soon/preliminary-documentation/connect-data/data-quality-mistakes-and-how-to-catch-them) and [Column Preprocessing](/kumo-engine/1.0.0/kumo-studio-guide-coming-soon/preliminary-documentation/connect-data/column-preprocessing) for the complete workflow.