Model Drift / Replacement

cancel
Showing results for 
Search instead for 
Did you mean: 

Model Drift / Replacement

This article explains how to change the model that a deployment is pointing to in DataRobot. (Note that the process explained here is not available for external deployments.)

Model predictions tend to degrade in accuracy over time. For this reason,  DataRobot provides an easy way to switch over previously built (and deployed) models to newly built models which ensures the models are up-to-date and accurate.

First, click the Deployments tab and locate your deployment.

Figure 1. Deployments pageFigure 1. Deployments page

Select the deployment and then use the Replace model functionality found in the actions menu (Figure 2).

Figure 2. Replace model actionFigure 2. Replace model action

Substitute model options

The displayed model Overview page provides a few options for substituting the model (Figure 3). You can:

  • import a downloaded DataRobot model
  • Import a custom model from model registry
  • paste the URL of the model itself

Figure 3. Replacement model optionsFigure 3. Replacement model options

Finding the URL of the Model

The last option is the easiest one. To find the URL itself, just click on a model from the Leaderboard. The URL on your browser will change and you can just copy that.

Figure 4. Model URLFigure 4. Model URL

The only caveat is that the model should not use any new features that were not already present in your existing model.

Replacement Reasons

Finally, you can choose why you want to replace your model and then click on Accept and replace to apply the change.

Figure 5. Model replacement reasonsFigure 5. Model replacement reasons

The REST API will now start using the new model and you can make predictions without further changes.

More Information

If you’re a licensed DataRobot customer, search the in-app platform documentation for Replace model action.

Version history
Last update:
‎07-13-2020 03:34 PM
Updated by:
Contributors