Work with feature lists¶

DataRobot builds models using feature lists—subsets of features from your dataset. DataRobot automatically generates several feature lists. You can also create custom feature lists, using domain knowledge to select the features that will be most useful in building accurate models.

Following are examples of feature lists DataRobot generates.

Takeaways¶

This tutorial shows how to:

• View feature lists
• Create new feature lists
• Compare models built with different feature lists
• Run Autopilot on a specific feature list

View feature lists¶

View the features in the lists that DataRobot generates automatically.

1. Import your dataset and select a target.

   The sample dataset featured in this tutorial contains patient data.

   

   The goal is to predict the likelihood of patient readmission to the hospital. The target feature is readmitted.

2. Scroll down to the Project Data tab.

   By default, the All Features list displays.

   

   This list identifies features DataRobot determines to be non-informative.

   

   In this example, some features have too few values and some are duplicates.

3. Click the Feature List dropdown menu and select the Informative Features list.

   

   The Informative Features list displays, and the non-informative features are removed.

   

Create a feature list¶

Select features to build your own custom feature lists.

1. In the Project Data tab, select features using the check boxes to the left of the feature names.

2. Click + Create feature list and enter the new feature list name to save your custom feature list.

   

Create a feature list from an existing list¶

Use the menu to select an existing feature list, then add or remove features to create a new feature list.

1. Click Menu on the top left of the Project Data tab and click Select features by feature list.

   

2. Add or remove features using the check boxes to the left of the feature names.

3. Click + Create feature list and enter the new feature list name to save your custom feature list.

Filter and select by var type¶

Filter and select features by variable data type.

1. Click Menu on the top left of the Project Data tab and click Select features by var type.

   

2. Add or remove features using the check boxes to the left of the feature names.

3. Click + Create feature list and enter the new feature list name to save your custom feature list.

Build and compare models¶

Compare models built with different feature lists by comparing optimization metrics. Choose metrics based on the project type, for example, regression, binary classification, or multiclass.

1. After loading your data and setting your target, click the Feature List dropdown menu and select a feature list.

2. Click Start to begin modeling.

3. When modeling is complete, click the Models tab at the top.

   

   The Leaderboard lists the generated models and indicates the feature list that was used to generate each.

4. In the Metric dropdown menu, select a metric to use to compare the models.

   

   This example uses LogLoss as the metric. For LogLoss, the lower the value the more acurate the model.

Rerun Autopilot on a feature list¶

After you build your models, you might decide to customize a feature list and generate more models.

1. Create a custom feature list.

2. On the Data tab, click the Feature Lists tab.

   

3. Click the menu to the right of the feature list you want to use to build new models and select Rerun Autopilot.

   

4. In the Rerun Modeling window, select the Modeling mode and click Rerun.

   

Learn more¶

Documentation:

• Explore feature list details.
• Assess feature impact—how important a feature is in the context of a particular model.