I am looking to implement CI/CD with DataRobot, is there any documentation related to that? Can I achieve CI/CD in trial version?
this a good python article with alot of examples and code samples
Great! Thank you for this. This can get me started. However, if you can help me with some template or a highlevel code sample for CI/CD that really will be helpful. Since I am relatively new to this DevOps concept I haven't given a thought.
Yes; although the SDK makes it fairly easy to do, one must be a bit comfortable with coding in Python. Note there are example jupyter notebooks as well to follow along/modify. Note the Predicting Bad Loans and Modeling Airline Delay project examples. You will also see these projects and models in the GUI if created with the API. As far as dashboards about the models; you can understand the models with tools like the ROC Curve, Feature Impacts, Feature Effects, etc. and can export images from the GUI (as well as data) - data can also be brought back to python and rendered as seen in the Advanced Model Insights example.
The Python SDK is not running DataRobot locally, but interacting with the same account you are using for the AI Trial in the GUI via your browser; so it is subject to the same expiration. Most of the heavy lifting is done by DataRobot and the SDK; Python is used for orchestrating tasks here, but deep expertise in Python is not required.
Thank you for the response. So, technically I will have to write complete code for me to implement CI/CD? Is there anyway I can showcase the models in a dashboard? What limitations will I be having if I go with the coding part?
Also, just for the sake of the code, if I use the Python SDK, can I continue with the trial version or will it get expired? And, what is the level of expertise should one be having to implement CI/CD on these?
You have the ability to interact with DataRobot programmatically via its API and a Python or R library that wraps it. Most of what you see in the GUI is available via it; the Python SDK documentation is available here. With it you can bring data into DataRobot, model it, and promote those models to registered deployments. The project creation to train a new model can be done manually, kicked off by a process outside DataRobot, or leveraging signals from DataRobot as well; for example, data drift being detected within the model monitoring for a deployment.
New model deployment can take the option of creating a completely new deployment, or simply replacing the model behind an existing deployment. In the case of the latter, the calling application may need no change at all; it will simply start receiving results from the newly deployed model. Note this assumes all necessary features are available and being sent in the same way; new features, or a change in preparing them, would require some updates in the way the requestor has prepared the data.