What does the process of exporting a model trained in DataRobot using AutoML look like? I am evaluating Data Robot for my team and we need to be able to export the model to deploy locally.
I see that there is an option to download a binary file after training. How can this binary file be used, and how does it compare to a .pickle file? (I understand the format is completely different, but I am interested in knowing whether I would be able to load it and use it in a similar way or does it need to be open with say a DataRobot API)
Thank you for your time!
I was just reading some of the documentation for exporting models, which might help you. https://docs.datarobot.com/en/docs/predictions/ui/model-transfer.html#exporting-models
please select Accept Solution if this is what you needed!
Thanks for the reply! I can't find whether the model can be loaded outside of DataRobot though. Should we assume it is not possible since it is not mentioned anywhere?
Just looking at the Download model packages section of the docs. The portable prediction server does allow you to run your model locally to get predictions. From my understanding the model packages (.mlpkg files) would work like a pickle file as you wouldn't have to do anything like retrain the model after downloading it.
And there is also the on prem installation of DataRobot if your just asking for security/reliability reasons. Why do you want to deploy locally?
Hi @lsavasta , from your query I presume you have found some model from the platform development approach worthy of your consideration for deployment. DataRobot takes similarly wholistic approach to deployment of models, both externally and internally developed. The deployment approach would obviously consider model degradation from accuracy and drift as well as monitoring, alerts, governance, etc.
We cover these in great detail in our MLOps Virtual Instructor-led classes listed here:
I hope you can see from the classes the comprehensive framework in which DataRobot enables models to be deployed and managed in both large and small environments with APIs.
In addition, DataRobot also allows models from the Leaderboard to be deployed via it's DataRobot Prime code and we cover that in our Foundation courses listed below:
You can also clarify your understanding of both the DataRobot platform's modeling and deployment approaches in the class and we look forward to seeing you in there.
Welcome to the DataRobot Community @lsavasta - happy to have you here
I see you received a couple of good responses / tips from the community. Did these help? (If so, please select Accept as Solution to let everyone else benefit from the right answer.) Need something more? Please let us know.
Looking forward to hearing more!