This post has been republished via RSS; it originally appeared at: New blog articles in Microsoft Tech Community.
Microsoft partners and customers have been building metrics monitoring solutions for AIOps, Business and predictive maintenance, by leveraging the easy-to-use time-series anomaly detection service: Anomaly Detector.
In this GA service, we provide two kinds of feature, Univariate Anomaly Detection (GA) and Multivariate Anomaly Detection (Public Preview). Those two features both have their stages to perform, and since last April when Multivariate Anomaly Detection is Public Preview, we've been engaging with more and more customers to support on their industries including energy, equipment health, manufacturing, etc.
What's new in this version?
We are excited to announce that we are adding more powerful capabilities in Multivariate Anomaly Detector(MVAD) today. In the latest version(v1.1-preview.1) of this API, we implement a new API in inference process, in a synchronous manner, which means you could get the anomaly detection results immediately once you call this API, this is a big change compared with previous inference process and will be more commonsensible and easy-to-use.
Also, we add a new field named 'interpretation' to give more explanation on an anomaly, like which variables have huge correlation changes and cause the anomaly. These updates will support you better leverage MVAD and get more useful information to analyze, then take actions.
Overview of this new API
With the synchronous API, you can get inference results point by point in real time, and no need for compressing and uploading task like training and asynchronous inference. Here are some requirements for the synchronous API:
- Need to put data in JSON format into the API request body.
- The inference results are limited to up to 10 data points, which means you could detect 1 to 10 timestamps with one synchronous API call.
- Due to payload limitation, the size of inference data in the request body is limited, which support at most 2880 timestamps * 300 variables, and at least 1 sliding window length.
In total, there are 8 APIs available for you to use in Multivariate Anomaly Detection, check them out in Anomaly Detector v1.1-preview.1 API Reference.
Overview of 'interpretation'
You will get a JSON response of inference results in real time after you call a synchronous API, which contains following new fields, and these fields will help you on root cause anlytics. Therefore, you could not only know when there's an anomaly, but also know which variables contributed to this anomaly, which will save your time of exploration on all variables.
| Field | Description |
|---|---|
| interpretation | This field only appears when a timestamp is detected as anomalous, which contains variables, contributionScore, correlationChanges. |
| correlationChanges | This field only appears when a timestamp is detected as anomalous, which included in interpretation. It contains changedVariables and changedValues that interpret which correlations between variables changed. |
| changedVariables | This field will show which variables that have significant change in correlation with variable. |
| changedValues | This field calculates a number between 0 and 1 showing how much the correlation changed between variables. The bigger the number is, the greater the change on correlations. |
Getting started is really simple
You have two quick options to get started:
Option 1: Follow the steps in How to: Use Multivariate Anomaly Detector on your time series data, it contains all the process for you to train a model and do inference, both in asynchronous and synchronous way, perfectly match for those who are new to MVAD and want to get some nitty-gritties of our service. This document also includes detailed codes and best practices.
Option 2: For those who want to hands on directly on code and familiar with SDK, this MVAD SDK Sample Notebook is perfect for you!
Once you have the materials ready, you are all set to experience these new capabilities!
Learn More
- Documentation about these new capabilities: Use Multivariate Anomaly Detector on your time series data.
- Keep updated on What's new of Anomaly Detector.
- Join our community for better support!