Predict performance degradation and service failure¶
Access this AI accelerator on GitHub
This notebook is designed to provide a predictive framework for managing and maintaining your machine learning models using DataRobot MLOPs. It focuses on preemptively identifying potential model performance degradation and service failures, enabling your organization to proactively address these issues before they adversely impact operations. This approach ensures the sustained efficiency and reliability of predictions that are integral to your business operations.
Early detection of model performance deterioration allows for timely interventions. These interventions could range from adjusting DataRobot's retraining policies to other corrective actions, ensuring the maintenance of optimal model performance. Similarly, predicting potential service infrastructure issues facilitates preemptive maintenance. This not only reduces downtime but also enhances service reliability and provides insights into the root causes of these issues.
The notebook demonstrates how to leverage DataRobot MLOps functionality to predict if a machine learning model is likely to degrade within a specific time period and if infrastructure failures may occur. It utilizes DataRobot's Python AI capabilities to collect various characteristics and metrics that DataRobot MLOPs tracks for your deployed model, thereby enabling the construction of a predictive model."
This notebook outlines how to:
- Establish a connection with DataRobot and access the relevant deployment details
- Create a training dataset
- Define a target for predicting model performance degradation
- Build a TS Model Degradation project
- Define a target for predicting Service Failure
- Build a TS Service Failure project
- Retrieve modeling results from the DataRobot Projects
- Augment the original deployment with custom metrics based on these predictions