8b. Refine Ideation
In the previous tutorial, you ran an Ideation in automated mode, where the system independently generated a variety of features and models.
In this tutorial, you’ll learn how to refine that ideation by incorporating domain expertise or leveraging prior results to produce more meaningful and efficient outcomes.
You will learn how to:
- Prune uninformative columns using prior feature selection results and enable filter detection during ideation configuration.
- Clone or reset an existing ideation to edit column transformations.
- Create a new feature selection and train new models.
Prune Uninformative Columns and Enable Filters¶
By default, Filter Detection is skipped because it can generate a large number of features. Before enabling it, first prune uninformative columns using the results from a prior feature selection to limit the number of generated features.
Once pruning is complete, you can run ideation up to the Filters stage to review and refine suggestions.
Steps¶
-
Start a new ideation Go to Ideation and click
to begin a new process.
-
Edit ideation details Update the ideation name and description by clicking
.
-
Configure ideation setup Click
to open the configuration panel.
Under Semantic Detection Setup, select a prior feature selection to prune uninformative columns.
In the Filter Setup section, uncheck “Skip Filters.”
-
Run ideation up to the Filters stage Set the auto-run configuration to stop at the Filters stage, then click
to start ideation.
-
Review suggested filters Examine the filters generated at the Filters stage.
-
Manage filters Click
to view, cancel, or create filters as needed.
Once your review is complete, either proceed to the next refinement step by clicking or move to the Modeling phase by clicking
.
Clone Ideation and Edit Column Transforms¶
Next, you’ll clone a prior ideation to modify column transformations.
-
Clone Ideation Navigate back to
Fully Automated
Ideation and open itsTransform
step. Clickto clone the ideation at the
Transform
step.
-
Review suggested transforms Choose a table of interest and examine the suggested transforms.
-
Manage transforms Click
to view, cancel, or create new transforms.
After completing your review, either proceed to the next refinement step or move to the Modeling phase as before.
Run New Feature Selection and Model¶
You can run new Feature Selections in two ways:
- Configure the
Feature Selection Setup
before starting a new ideation. - Run a new Feature Selection from a completed ideation.
In this section, you’ll create a new selection from an existing ideation.
-
Navigate back to
Fully Automated
Ideation and open itsFeature Ideation
step.
-
Click the Magic Wand
to start a new Selection.
-
Go to the
Rule Based
tab and configure the following:- Feature Selection Name = Top 1 per theme
- No. of Top Features Overall = 500
- No. of Top Features per Theme = 1
- Selection Logic = AND
-
Review selected features Once the selection completes, review the chosen features.
-
Train a new model Go to the
Modeling Training
step and clickto train a model using the new selection.
-
Review model performance Check the results in the Ideation Leaderboard. As expected, this simplified selection may perform poorly.
Add Features to the Feature Catalog¶
Features used in Ideation models are automatically added to the Catalog. You can also manually add additional features.
-
Go to the
EDA
step and itsFeatures
tab. Clickand clear all filters and searches using the clear icons.
-
Select the
NEW
Readiness filter (features not yet in the Catalog).
-
Choose a few features (for example, two).
-
Click
to save them to the Feature Catalog.
-
Confirm the save by clicking
.