Skip to content
For Self-Managed AI Platform users running v11.1, see the on-premise platform documentation
NextGen UI documentation > Compliance > Audit model code

Audit model code

For the purpose of auditing and compliance, among other things, there are a variety of entry points for accessing model code. The major options are listed below:

  • Model compliance documentation can be generated for every model created in DataRobot. Compliance documentation summarizes all of the components of the model and validates that it is conceptually sound and is designed specifically to support any required audit or Messaging Records Management (MRM) compliance processes.
  • DataRobot model blueprints provide a graphical representation of each preprocessing step, modeling algorithm, and post-processing step that goes into building a model. Blueprints are customizable, allowing users to edit out-of-the-box blueprints using built-in tasks and/or custom Python or R code.
  • Every step (node) within a DataRobot blueprint is documented in the model documentation, which can be accessed by clicking the task within the blueprint task.
  • Full end-to-end code is available for some (although not all) DataRobot blueprints:
    • Eureqa Models are proprietary algorithms that construct models that balance predictive accuracy against complexity. The output of an Eureqa model is a human-readable and interpretable analytic expression, which is easily reviewed by subject matter experts.
    • Rulefit models create an ensemble of simple rules and coefficients for each rule, which can be exported as Python or Java code.
    • GAM and GA2M models provide the ability to create and download rating tables of the model's complete, validated parameters. Rating table validation ensures that the downloaded parameters are correct and that the model's performance outside of DataRobot can be reproduced. (Use this, for example, to visualize how changing the state impacts the score. For each variable, the binned ranges show the minimum and maximum possible values that variables can influence, so prediction explanations become deterministic.)
  • DataRobot also provides the ability to export models as scoring code or as a portable prediction server, allowing execution of prediction jobs outside of the platform.