- Support for ONNX resources:
- Added REST API for management of pre/post-processing resources for ONNX models.
- Updated UI to enable ONNX resource management.
- Support for ONNX pipelines:
- Added REST API for management and execution of ONNX pipelines.
- Updated UI to enable ONNX pipeline management and execution.
- Enhanced PMML batch processing results:
- Provides summary statistics for score matching tests.
- Available for single model and model group apply.
- Support for ONNX models:
- Dedicated microservice which allows deployment and management of ONNX models.
- Updated UI to enable ONNX model management and inferencing.
- Support for Multi-Variate State Space Models:
- Representation and computation of confidence intervals as an output for state space models.
- Support for multi-valued output for multi-step forecasts and confidence intervals.
- Enable grouping of multiple models:
- Models with homogenous input/output schema can be added to a logical group.
- Inferencing can be done on a model group – one or more models at the same time.
- Grouping also allows combining multiple versions of same model into a logical group.
- Performance improvements to the job scheduling feature by including pagination.
- Enable ingestion of time series data:
- Given any time series data, users can generate ARIMA models.
- This functionality will be backed by REST endpoints offering users the option to invoke them from any client.
- Enhanced support for time series:
- Representation and computation of confidence intervals as an output for time series models.
- Added support for Kalman Filters for the “exactLeastSquares” prediction method.
- Modification of charts for scheduled data processing:
- The histogram is replaced with a line chart.
- Timestamps pertaining to every measurement is part of the line chart.
- Support analysis of device data using scheduled jobs.
- Schedule batch jobs for processing measurements from devices or device groups against an available model.
- The job scheduler can be used to trigger one-time or periodic jobs on data captured from devices.
- The scheduler allows you to map device data to model inputs by providing a mapping tool. Periodic executions of batch jobs can be useful when aggregate information on model’s predictions is required for a desired time period.
- Support for Object Detection using RetinaNet.
- Support for univariate seasonal ARIMA time series models with “conditionalLeastSquares” prediction method.
- Compliance with PMML 4.4 release features:
- Support for Lag@aggregation=stddev
- Support for Constant@missing
- Support for Anomaly Detection release schema
- Embedded Zementis engine within a microservice package for Cumulocity IoT.
- Minimized footprint for the Zementis microservice along with improved performance.
- Comprehensive REST API for model and resource management and data processing.
- Intuitive web application which leverages the REST API of the Zementis microservice.
- Added support for following capabilities -
- Upload/delete/download of PMML models and custom resources to Cumulocity IoT. Users can also upload compressed PMML file (ZIP). The PMML file needs to be the only entry in the ZIP file.
- Enabled batch processing of data against available models on Cumulocity IoT. Users can even upload compressed CSV/JSON/Image files (ZIP) for batch processing.
- Visualize Model KPIs including memory metrics and prediction metrics. These metrics help users to view the memory footprint of every model that they have and also summarize the real-time predictions from each model.
- Model activation/de-activation at runtime. De-activation evicts the model from JVM heap, allowing the Zementis Microservice to manage more models than available heap.
- Support PMML 4.3 standard.
- Added compliance with upcoming PMML 4.4 release features:
- Support for multiple model methods “weightedSum” and “weightedMedian”
- Support for new MiningField@invalidValueReplacement attribute
- Support for new Lag@aggregate attribute
- Support for new built-in Functions: isValid, isNotValid, modulo(%), sin, asin, sinh, cos, acos, cosh, tan, atan, tanh, rint, hypot, ln1p, expm1