industrial-edge issues

Support foreground deletion

1

Introduce data drift

We should showcase data drift detection and how we can detect and maybe retrain the model based on updated anomaly data.

mamurak

Actually create a meaningful model update

Currently the anomaly detection model is classifying anomalies well enough in the context of the line dashboard app. How about the initial model doesn't perform well, so we see how...

mamurak

Model candidate selection

Apply advanced routing schemes (e.g. shadow scoring) for model candidate selection.

mamurak

Incorporate hyperparameter optimization

In a more realistic scenario, the data scientist would run hyperparameter optimization during model training. We should incorporate this into the model CI/CD pipeline.

mamurak

Create SDK version of ML CI/CD pipeline

As an alternative to the current Elyra implementation of the ML CI/CD pipeline, we should prepare an implementation via the KFP SDK, which allows us to leverage more advanced features...

mamurak