The model's Downloads tab allows you to:
- Export Models (1): Create a binary file of the model, which you can then use to transfer the model to a standalone prediction server or your Hadoop cluster. Not available for Managed AI Cloud deployments.
- Download Scoring Code JAR (2): Creates an exportable JAR file for tree-based and linear models (indicated by a SCORING CODE icon () on the Leaderboard).
- Download Exportable Charts (3): Download a single ZIP archive containing charts, graphs, and data for that model. Charts and graphs are exported in PNG format; model data is exported in CSV format.
Contact your DataRobot representative for information on enabling Scoring Code JAR capabilities.
Exporting models for transfer¶
You can use the Downloads tab as part of the process for transferring models from the DataRobot application where they were built to another cluster. You might want, for example, to export a model from DataRobot and then import it to a standalone prediction server environment. Click the Download Model link of Model Export to create and download a binary file for model transfer. If prompted, save the resulting file(s) locally.
For full details, see the section on model transfer.
Model import requires Prediction Admin permissions. Both import and export are not available for Managed AI Cloud deployments.
Generating scoring code¶
DataRobot automatically runs code generation for those models that support it, and indicates code availability with a SCORING CODE icon on the Leaderboard. This option allows you to download validated Java scoring code for a predictive model without approximation; the code is easily deployable in any environment and is not dependent on the DataRobot application. Once exported, you can view the model's source code to help understand each step DataRobot takes in producing your predictions.
For full details on using the downloaded scoring code, see the section on Java code generation.
Scoring code or DataRobot Prime?¶
Both DataRobot Prime and the Scoring code JAR option provide code that is easily deployable for making offline predictions, independent of the DataRobot application. When deciding which approach best suits your needs, consider the following:
|Requirement||DataRobot Prime||Scoring Code JAR|
|Accuracy to original model||Creates code that approximates the original model.||Creates the exact same predictions as the original model.|
|Transparency||Code is a simplified version of the model, creating "business rules" for predictions; use this when you do not need an exact model replica.||Code is a more complex but exact replication of the model; use this when you need to see every step DataRobot uses for making predictions with the model.|
|Benefits||Allows you to more easily view and understand the a simplified version of the model.||Provides complete transparency for model predictions.|
For full details, see the reasons to use DataRobot Prime.
Click the orange Download link to download a ZIP archive of model information. To save individual charts and graphs, use the Export button, described in the section on exporting. The archive contains the following model types:
|Feature Effects *||✔||✔|
|Feature Fit *||✔||✔|
|Variable Effects (Coefficients)||✔||✔|
|Tree-Based Variable Importance||✔||✔|
* If computed, you can export the chart image for individual features. If you choose to export a ZIP file, you will get all of the chart images and the CSV files for partial dependence and predicted vs actual data.