0
3.0.1
Vancouver
TSI Detect is an AI capability that offers multivariate (multi-input) anomaly detection on time series data. Typical uses of such capability are in the AIOps domain. With multiple systems and hundreds of metrics per system, we can't rely on hand-crafted rules to correlate metrics or extract patterns. That's why TSI Detect comprise unsupervised Deep Learning models to learn the “normal behavior” of your data, and is followed by a classifier (when labels, aka user feedback, are available) to classify anomalies as relevant/useful or not, that improves over time with the user feedback. This capability will help prevent incidents, making it faster to recover from any incidents and maintain client satisfaction.
- Filtering out noise resulting from short-lived, single-metric anomalies.
- Capture more complex (collective) anomalies missed by univariate model, meaningful in the context of multiple metrics, by having a deep learning multivariate time series model.
- Learning without labels by capturing what the typical profiles are from the raw measurements.
- Performance improvement over time given user feedback (continuous learning).
- Ability to control the model sensitivity, to receive more or less anomalies.
- Ability to accommodate basic user-defined rules to refine and filter anomalies, for example, the minimum anomaly duration.
- Measures of anomaly severities and fundamental root cause analysis (i.e., what input metrics contributed to a particular anomaly in a given time window).
- Live operation in real-time (i.e., searching for anomalies with a defined time granularity) or on-demand (i.e., another system can query if there are/were anomalies in a configurable time window).
- New
- CIs are now clustered by CI type and metric type, and one model is trained per cluster, instead of per CI.
- Scheduling for batch prediction request jobs
- Changed
- Create re-training request jobs before scheduled job "TSI Detect Schedule Training Requests"
- Reduced the amount of logs
- Fixed
- Fix misalignment of anomaly results between batch and cascade predictions
- Fix calculation of anomaly score/attributions for cascade predictions
Required plugins:
- com.glide.hub.designer
- com.glide.platform_ml
- com.snc.clotho