Skip to main content
You create a feature table to make raw data consumable by machine learning models during training. This task walks you through creating a feature table in NexusOne by selecting source tables, a target table, and labels.

​
Prerequisites

  • Appropriate permission: nx1_automl_admin and nx1_crew or nx1_automl_user and nx1_crew.
  • An existing domain created using DataHub.
  • A previously ingested dataset in the domain.

​
Create a feature table from prediction problems

Perform these steps to create a feature table from prediction problems.
  1. Log in to NexusOne.
  2. On the top navigation bar, hover your mouse over AI and then select AutoML > Feature Engineering.
  3. Click New Proposal.
  4. Define a target training table by doing the following:
    1. Select a DataHub domain.
    2. Select a schema and table in that domain.
  5. Describe your prediction problem using natural language.
  6. Define one or more source tables by doing the following:
    1. Select a catalog.
    2. Select one or more schemas and table names in the catalog.
  7. Select grain columns. These are the identifier columns that uniquely identify each training row.
  8. Choose one of the following label sources that describes how NexusOne obtains the label for model training:
    • Existing column: It uses a column in your source data as the label. If you’re using this option, then select a label column.
    • Natural-language deviation: It uses a natural language description about how to derive the label from existing source table columns. If you’re using this option, then enter a description.
  9. Enter a positive class value. It represents the outcome the model should treat as the positive prediction and it only applies to binary classification problems.
  10. Optional: Enter a SQL join hint to specify table joins.
  11. Click Submit to create the feature table.

​
Additional resources

For a high-level overview of feature tables and how they’re used in NexusOne, refer to Feature engineering.