DKube 3.0 is now in production, and includes important new capabilities:
DKube Monitor enables the production engineer to ensure that a deployed model continues to provide acceptable inference results as the input data or business goals change over time. Alerts can be set up based on measurement tolerances, and the results are viewed through an intuitive dashboard. Deviations can be quickly identified and resolved through retraining and redeployment.
This capability is as important as the initial model development, since the performance of deployed models will generally degrade over time, and the results need to be identified and improved before they impact business outcomes.
These new features add to the leadership capabilities DKube users have come to expect:
Machine learning has been traditionally focused on the important initial training phase. This ensures that a model will provide the best outcomes for the expected live data stream - at least for a while. However, it is well known within the machine learning community that however good a job you do at training a model, your results will
How to launch hyperparameter tuning runs taking advantage of the Katib based hyperparameter tuning available in Kubeflow. Pick the best model from the multiple training runs as the winner based on pre-set criteria.
How to launch a single model training or data preprocessing run through the DKube UI while passing new parameters to pass to the training program via the UI and designated environment variables. Learn how to track the performance characteristics and the lineage of the run with the dataset and model versions used or created in the process.