From my understanding, DataRobot will not be able to use continuous AI automatic retraining, because one assumes that these models are trained models monitored by MLOps. For instance, the python models are pickled (serialized) before being put in out MLOps and monitored. I cannot see how automatic retrain will work. However, you can change the model with another model.
Thanks @dalilaB, Both MLDev and custom Inference models are trained models before being monitored by MLOps, I assume then what is passed to the model registry just contains model information (without the code used for training that model ect)?
I assume it could be done with a parameter of some kind allowing DR to pass training data to the model, if the model registry had that ability.
One solution is to register the model in our model registry and then you can have it retrained, which is what is drawn in the image above. Still, I will reach out to our MLOps experts and get back to you on Monday.
Hey @IraWatt I'm sure you've looked here already but just in case: https://community.datarobot.com/t5/blog/what-s-new-in-datarobot-release-7-2/ba-p/12469. See the Continuous AI (public preview) info/demo video.
Hi Ira, I am from former Algorithmia. We got acquired about 2 Month ago by DataRobot. The custom inference capability brought by Algorithmia differ quite a bit from the vanilla offering.
ModelCatalog, Versioning, Pipelining, Governance, Languages, Dependencies, Compute, Orchestration, Hardware, Monitoring are features provided across models built in R, Python, Ruby, Rust...DataRobot, H2O, Dataiku, AWS SM...etc using autoscaling capabilities running Kubernetes on Dockers, deployed on premise or in the Cloud.