Using darts python library to create a custom Saved Model
I've followed the tutorial here:
Importing serialized scikit-learn pipelines as Saved Models for MLOps - Dataiku Developer Guide
and I've been able to develop a model using the darts==0.30.0 library, having wrapped it in the standard scikit-learn pipeline. My issue is with the very last part of step 3 of this tutorial where I want to deploy_run_model but I'm unable to because my environment was missing the Visual ML packages. When I added those packages to the code env, there are many issues with mis-matching dependencies wrt the darts package, so I attempted to downgrade the darts lib but now am stuck waiting for a 2 hr environment update.
Is there any plan to update the versions of standard Visual ML packages?
Operating system used: Windows 11
Best Answer
-
Turribeach Dataiku DSS Core Designer, Neuron, Dataiku DSS Adv Designer, Registered, Neuron 2023 Posts: 2,166 NeuronThe Visual ML package sets you get when you add them in a Dataiku code environment are dependent on the Python interpreter version your code environment has. For instance in a Python 3.7 code env Dataiku will add "scikit-learn>=0.20,<1.1" but in a Python 3.10 code env Dataiku will add "scikit-learn>=1.0,<1.6" and in a Python 3.11 code env Dataiku will add "scikit-learn>=1.1,<1.6". So in other words you should use a newer Python version in your Dataiku code environment if you want a newer scikit-learn.
Answers
-
Thanks, it seemed to work when I downgraded darts to 0.20.0 and used the Visual ML package set targeted for a Python 3.10 code env
-
Correction: needed to upgrade to darts==0.23.0 and then xgboost==1.6.0. Also needed to add mlflow=2.13.0. Then it worked.