Skip to content

Click in-app to access the full platform documentation for your version of DataRobot.

Feature lists

Feature lists control the subset of features that DataRobot uses to build models. You can use one of the automatically created lists or manually add features from the Data page or the menu. You can also review, rename, and delete (some) feature lists. The list used for modeling is called the default modeling feature list. That is, it is the feature list selected when you clicked the Start button.

If you don't override the selection, DataRobot uses either of the following lists to build models:

  • all features that provide information potentially valuable for modeling (the Informative Features list).
  • all features that provide information potentially valuable for modeling with any feature(s) at risk of causing target leakage removed (the Informative Features - Leakage Removed list).

You can select features to create a new feature list, before or after EDA2. The target feature is automatically added to every feature list. Once created, the new list becomes available in the Feature List dropdown. DataRobot highlights the active list, which controls the display of features on the page, in blue.

Note that the Project Data tab defaults to showing All Features, which is not actually a feature list but instead a way to view every feature in the dataset.

Select a feature list

To use a feature list other than the list assigned by DataRobot, select the list to use as the default modeling list from the Feature List dropdown. The new setting is reflected under the Start button:

Create feature lists

If you do not want to use one of the automatically created feature lists, you can create customized feature lists and train your models on them to see if they yield a better model. You can create these lists from the Data page or the menu. Additionally, you can create lists based on feature impact from the Feature Impact tab, including lists with redundant features removed. You can later manage these lists from the Feature Lists tab.

Create feature lists from the Data page

To create feature lists from the Data page:

  1. Select the Project Data tab.
  2. Optionally, from the Feature List dropdown select All Features to display all columns (features) in your dataset.

  3. Use the checkboxes to the left of a feature name to select a set of features. When you select the first feature, the Create Feature List link becomes active.

  4. Select each feature you want added to your new list and click Create Feature List.

    Enter a name in the resulting dialog box and click Create feature list. The page display updates to show only those features that are part of the new list (highlighted in blue in the Feature List dropdown).

Tip

Click in the box to select all, or deselect any, selected features.

Create feature lists from the menu

You can use the Menu options to quickly select features for a new feature list. Click the Menu to expand:

Clicking a feature list name causes DataRobot to select all features on the displayed page that are members of the chosen feature list (set by the Feature List dropdown). For example, set the Feature List to Informative Features and then, from the menu dropdown, select the example created above (Top5). DataRobot automatically selects (checks the left-hand boxes) of the five features in the Top5 list. You can now use that as a base and add or drop features to create a new list:

Add the new features, name your list, and click Create. The new list is available for selection across the project (from the Feature List dropdown).

Feature Lists tab

The Feature Lists tab of the the Data page provides a mechanism for managing feature lists. It provides a summary (name, number of features, number of models, created date, and description) of DataRobot-created and custom feature lists and allows you to delete or rename (some) lists to help avoid clutter and confusion. A lock() next to the name indicates the list cannot be deleted.

After building models, the list includes additional automatically created lists (1) as well as any custom lists (2):

Manage feature lists

DataRobot provides several tools for working with feature lists. Depending on how the list was created (automatically by DataRobot or manually by a user), or whether it has been used to create models on your Leaderboard, the actions may behave differently:

The following table describes the actions:

Icon Description
Exports features that are part of the selected list as a CSV file.
Opens the selected feature list on the Project Data tab.
Provides a dialog to let you edit the list name and/or description. (Automatically created feature lists cannot be renamed although the description can be changed.)*
Restarts Autopilot using the selected feature list.*
or Deletes the selected list (or indicates it cannot be deleted). Automatically created feature lists cannot be deleted.*

* You must have User-level or above project access to delete or rename feature lists, as well as to restart Autopilot.

Tip

You cannot add or remove features from a feature list. Instead, create a new feature list with all desired features.

Edit feature list names and descriptions

When creating a custom feature list, you simply name the list in the initial dialog. From the Feature Lists tab you can append a description to the list. To add that description, or edit an existing description, highlight the list and click the pencil icon ().

You can change a description, but not a name, for a DataRobot-created list.

Rerun Autopilot

You can launch a re-run of Autopilot from the Feature Lists tab by clicking the retrain icon (). Clicking the icon launches a dialog; select Restart Autopilot to rebuild the project with the new list.

  • If you restart while models are building for the project, DataRobot halts the feature list that is currently running (i.e., stops building new models with it) and restarts Autopilot, from the beginning, using the selected list.

Note that this is the same action as rerunning Autopilot from the Configure modeling settings link available in the right-panel Worker Queue.

Delete feature lists

Deleting a feature list also deletes any models in the project that were built with that list. Only custom feature lists can be deleted (no next to the name). If you click to delete a custom feature list that has been used for modeling, DataRobot warns with the number of models impacted:

You cannot use the delete function if the feature list is:

  • an automatically created list.
  • the default modeling list for the project.
  • configured as a monotonic constraint feature list for the project.
  • used as the input feature list to create the modeling dataset for a time series project.
  • used in a model deployment (the model and its feature lists cannot be deleted until after the deployments are deleted).

Automatically created feature lists

DataRobot automatically creates several feature lists for each project. Note that:

  • time series feature lists differ.
  • features created from a search for interactions result in different lists (appended with a plus (+) sign).
  • a project's target feature is automatically added to every feature list.

The following describes the automatically created feature lists, although not all lists apply to a project.

  • DR Reduced Features: A subset of features, selected based on the Feature Impact calculation of the best non-blender model in the Leaderboard. DataRobot then automatically retrains the best non-blender model with this DR Reduced Features list, creating a new model. DataRobot compares the original and new models, selects the better one, and retrains this model at a higher sample size for model recommendation purposes. DR Reduced Features, in most cases, consists of the features that provide 95% of the accumulated impact for the model. If that number is greater than 100, only the top 100 features are included. If redundant feature identification is supported in the project, redundant features are excluded from DR Reduced Features.

  • Informative Features - Leakage Removed: The default feature list if DataRobot detects target leakage. This list excludes feature(s) that are at risk of causing target leakage and any features providing little or no information useful for modeling. To determine what was removed, you can see these features labeled in the Data table with All Features selected.

  • Informative Features: The default feature list if DataRobot does not detect target leakage. This list includes features that pass a "reasonableness" check that determines whether they contain information useful for building a generalizable model. For example, DataRobot excludes features it determines are low information or redundant, such as duplicate columns, a column containing all ones or reference IDs, a feature with too few values, and others. Informative features are sorted to the top of the Features list.

  • Raw Features: All features in the dataset, excluding user-derived features and including those excluded from the Informative Features list (e.g., duplicates, high missing values).

  • Univariate Selections: Features that meet a certain threshold (an ACE score above 0.005) for non-linear correlation with the selected target. DataRobot calculates, for each entry in the Informative Features list, the feature’s individual relationship against the target. This list is not available until EDA2 completes.

While not a feature list (not available for use to build models), the All Features selection sets the Project Data display to list all columns in the dataset as well as any additional transformed features.

Data page informational tags

The Data page displays tags to indicate a variety of information that DataRobot uncovered while computing EDA1. For example:

Tag Description
Duplicate A feature column is duplicated in the ingest dataset.
Empty Column contains no values.
Few values Too few values, relative to the size of the dataset, for DataRobot to extrapolate meaningful information from the feature. Not an indicator of the number of unique values, but instead domination of a single value, making the feature inappropriate for modeling. Specifically:
  • a numeric with no missing values and only one unique value.
  • a variable in which >99.9% is the same value
Too many values Too many values, relative to the size of the dataset, for DataRobot to extrapolate meaningful information from the feature. For categorical features, the label is applied if: [ number of unique values ] > [ number of rows] / 2 |
Reference ID Column contains reference IDs (unique sequential numbers). For numerical features, the label is applied if: [ number of unique values] = [number of rows]
Associated with Target Column was derived from target column.
Target leakage Indicates a feature whose value cannot be known at the time of prediction.

Updated October 26, 2021
Back to top