Monocle

Monocle is an MLOps platform for managing the full lifecycle of machine learning and artificial intelligence models, from development and experimentation to training, testing, tuning, deployment, and production performance monitoring. Monocle simplifies the historically time-consuming and labor-intensive tasks associated with integrating and maintaining machine learning models in operational systems. ATA addresses the challenges to productionizing machine learning by recognizing that data engineering and operationalized data science are a specialized form of software development that should embrace the workflows typical of data science along with the robust procedures of modern software delivery such as continuous integration and continuous deployment (CI/CD) and test-driven development (TDD).

Monocle Capabilities

  • Model registry to track experimentation and store model artifacts alongside training parameters and performance metrics.
  • CI/CD automation to support the software delivery lifecycle and the automatic wrapping of models into web services.
  • Model monitoring services to support the collection, analysis, and alerting of model performance metrics including accuracy, execution latency, memory usage, data distribution drift, and more.

Use Cases

  • Monocle has been used to integrate an object tracking algorithm into a streaming data pipeline to track the positions of aircraft in real time.
  • Monocle was used to train and tune a machine learning model that automatically identifies anomalous flight patterns. Its Model Registry allows data scientists to tune models by comparing hundreds of model runs and mesh grid hyperparameters.
  • The model monitoring service is used to track the performance of an aircraft anomaly model and the distribution of its input data, and will alert the ATA development team to sudden changes in overall aircraft behavior requiring retraining the machine learning model. Such changes can be caused by systematic shifts in flight patterns such as changes to flight regulations, airspace configurations, climate, and pilot training.
  • Models deployed by Monocle are monitoring the real time depth of hundreds of streams across the state of Virginia to alert emergency management of active or impending flood conditions.