Improve integration of custom and plugin models in VisualML
Antal
Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Neuron, Dataiku DSS Adv Designer, Registered, Neuron 2023 Posts: 97 Neuron
The functionality to continuously expand VisualML with plugin models, either from Dataiku or by coding our own plugins is very valuable to us. For example, we use the General Linear Models plugin a lot to support our actuarial team.
There are some areas where the integration of such plugins is not yet ideal in VisualML that seem like they should require low effort to solve.
- Enable bayesian hyperparameter optimization: currently only grid search is supported which can take a long time with large parameter sets. Extending this to allow parameter ranges to be set in the algorithm configuration instead of discrete values, like it is implemented for out-of-the-box models, would also be very helpful.
- Improve/support the generation of documentation for plugin models: the support for this is currently very limited. Also extending this to imported mlflow models would also be very helpful.
Comments
-
Antal Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Neuron, Dataiku DSS Adv Designer, Registered, Neuron 2023 Posts: 97 NeuronAlso, export model to deploy to Databricks (Unity Catalog).
Since they need to be scikit compatible anyway, this should be possible as well(?).