Skip to content

8b. Refine Ideation

In the previous tutorial, we explored the Automated Mode of Feature Ideation, where the system independently generated a comprehensive set of features.

Now, we focus on Semi-Automated Mode, which introduces an interactive layer to the ideation workflow. This mode empowers you to review, refine, and enhance the system's recommendations step by step, ensuring that the features align with your specific requirements and domain knowledge.

In this tutorial, you will learn how to:

  • Incorporate custom transformations, such as embedding UDFs, to enrich feature engineering.
  • Review and adjust the system's suggestions, from table selection to filters.
  • Understand how Semi-Automated Mode balances automation efficiency with manual refinement flexibility.

Step 0: Add New User Defined Function (UDF) to the Catalog.

Before starting a new Feature Ideation, we will register an "embedding" UDF that leverages the Sentence-BERT (SBERT) transformer model. This UDF will be used to transform the Product Group column of the PRODUCTGROUP table into embeddings.

  1. Navigate to the User Defined Function Catalog under the 'Formulate' section of the menu. Name

  2. Create the "embedding" UDF by clicking New UDF. Name

  3. Confirm that the new UDF is registered and visible in the catalog. Name


Step 1: Create New Feature Ideation

  1. Navigate to Feature Ideation under the 'Ideate' section of the menu and click New Ideation Button to start a new ideation process. Name

  2. Edit the Feature Ideation name and description by clicking Edit Button. Name


Step 2: Start Semi-Automated Mode

  1. Begin the workflow by clicking Next Step Button.

  2. After complete, the table selection results will be displayed for review.

Name


Step 3: Review Table Selection

  1. Click Name to view detailed table selection results. Name

  2. To open the detailed report:

    • Click Report Button next to the Ideation run name "Semi-Automated Mode".
    • Then click Report tab Button.

    Name

  3. Return to the table selection screen and proceed by clicking Next Step Button.


Step 4: Review Column Semantics Detection

  1. Review the Column Semantics Detection results. Name

  2. Click Name to view the report. Name

  3. Adjust semantic tags as needed. For example, assign the Gender column a semantic type under categorical/nominal_categorical/demographic_attribute/gender in FeatureByte's ontology. Name Name

  4. Click Next Step Button to continue.


Step 5: Review Transforms Detection

  1. Once the Transforms Detection process finishes, review the results for the INVOICEITEMS table. Name

  2. Click Name to view the report. Name

  3. Attempt to create a new Transform, 'Total_Cost / Quantity'.

    • Open the Transform window by clicking Name. Name
    • Click Name.
    • Select the 'Ratio' operation. Name
    • Choose the Total_Cost column as the numerator and the Quantity column as the denominator Name --- Name --- Name
    • Generate a name and relevance by clicking Name. Name
  4. Review the relevance explanation. If it is low (e.g., redundant with the existing column Unit Price), delete the transform. Name Click Name to delete the transform.

  5. Close the Transform window by clicking Name.

  6. Click Next Step Button to proceed.


Step 6: Review Filters Detection

  1. Once the Filters Detection process finishes, review the results. In this example, no filters have been detected. Name

  2. Create a new filter. Click Name for the INVOICEITEMS table. Name

  3. Select Filter Column. Choose Product Group as the filter column. Name

  4. Complete the filter condition by specifying the filter values. This will open a new windown listing all elligible values. Name Name

  5. Identify the most relevant values by clicking Name. Name

  6. Create automatically meaningful groups of values by clicking Name. Select one group if any is relevant. Name

  7. Finalize your value selection. Name

  8. Generate filter name and relevance. Name

  9. Check the relevance of the new filter. Name

  10. Save the new filter by clicking Name. Name

  11. Click Next Step Button to proceed.


Step 7: Review Feature Ideation Setup

  1. Review the suggested setup. Name

  2. Go to 'User Defined Function' section and click Name to use the "embedding" UDF to transform the Product Group column of the PRODUCTGROUP table into embeddings. Name

  3. Click Auto Run Button to complete Feature Ideation up to Feature Selection.


Step 8: Review the Feature Ideation Report

After the process completes, a feature selection will be displayed for your review.

Name

Accessing the Detailed Report: To view the full report describing each step of the ideation process, click Report Button next to the Ideation run name "Semi-Automated Mode".

Name

To visualize the full report with an indexed view in a new tab, click Report tab Button.


Step 9: Add Features to the Feature Catalog

  1. Go to the Features tab. Clear the search (if you used it) and any prior selection (if any) by clicking Clear Button
  2. Select features in the feature list by clicking Select All Button.
  3. Save the selected features into the Feature Catalog by clicking Save Feature List. Call the feature list "SHAP selection with embedding". Name

    Name


Step 10: Run Rule-based Selection

  1. Change to All features by setting the dropdown list to All features next to the Magic Ward. Name
  2. Start Feature Selection by clicking on the Magic Ward Magic Ward.
  3. Select the Rule-Based mode. In this example, we want the top feature for each theme if it is part of top 100 features overall. Name

Once the selection is complete, it is added to the selections list.

Name


Step 11: Add Rule-based selection to the Feature Catalog

  1. Go to the Features tab. Clear the search (if you used it) and any prior selection (if any) by clicking Clear Button
  2. Select features in the feature list by clicking Select All Button.
  3. Save the selected features into the Feature Catalog by clicking Save Feature List. Call the feature list "Top 1 per theme". Name