Experiment reference¶

Currently Workbench supports binary classification and regression projects. Because development is ongoing, see the release notes for a full list of supported capabilities.

When you create experiments in Workbench, you can access them in DataRobot Classic via the project management center.

Workbench always runs Quick Autopilot, whether on the initial run or when rerunning.

When first creating an experiment, the only feature lists available are Raw Features and Informative Features. After running models, additional automatically created feature lists may appear (for example, Reduced Features and Univariate Selections). If you want to model on a feature list you create, you cannot currently create the list in Workbench. You can, however, open the experiment from the project management center in DataRobot Classic and create a feature list there. When you re-open the experiment in Workbench, the feature list will be available.

When Workbench prepares a model for deployment, it trains the model on 100% of the data. While the most accurate was selected to be prepared, it was selected based on a 64% sample size. As a part of preparing the most accurate model for deployment, Workbench unlocks Holdout, resulting in the prepared model being trained on different data from the original. If you do not change the Leaderboard to sort by Holdout, the validation score in the left bar can make it appear as if the prepared model is not the most accurate.

You cannot currently create a blender model in Workbench; however, there is a workaround. To add a blender to your use case:

1. Create an experiment and build models.

2. Open the project in DataRobot Classic and blend models.

3. Reopen the experiment in Workbench.

The blender will be present in the Leaderboard.