Using the Parameterized Batch Scoring Command Line Scripts
(Updated October 2020)
DataRobot's in-app Platform Documentation identifies some Command Line Interface (CLI) scripts that can be used to create local file batch prediction scoring jobs. These scripts, found by searching the documentation for Batch Predictions Scripts, are available in both Python (generally for Linux and MacOS) and PowerShell (for Windows).
(Note that the deprecated script “Batch Scoring Script” may show up in search results as well. Make sure you ignore that topic and instead select “Batch Predictions Scripts.” If you are a DataRobot Managed AI Cloud user, you can view this direct link to access the documentation for Batch Predictions Scripts.)
When viewing that page of the in-app Platform Documentation, make sure you pay attention to the important instructions at the top of the page. As explained, when using a Python file you need to toggle the executable flag. Then, in either environment, the executable can be placed within the environment path so that it can be run from any location. The scripts themselves are downloadable from the documentation page.
You can use these scripts instead of using the Integrations tab code for Batch usage found in DataRobot. For a specific deployment, select Predictions -> Prediction API -> Batch (toggle); this provides a local file with an example batch scoring script.
All necessary values (Deployment ID, API Token, etc.) for the example script are hardcoded into the file. This is in contrast to the CLI scripts for download (mentioned above), which are generic and will expect these same values as command line inputs. When dealing with large volumes of deployments or automated tools, generally the parameterized CLI version that leverages a single common script is preferred for ease of maintenance and use. The deployment code (from the Integrations tab) is still a great reference for sourcing the values necessary to construct the executable call from the command line.
The command line utility requires at least three parameters and an API token; otherwise, the default values for all others parameters are appropriate.