17:57Anomaly Detector v1.0 Best Practices
From the creator
In our previous episodes of the AI Show, we've introduced to you Azure Anomaly Detector in both hosted cloud APIs and containers (Introducing Azure Anomaly Detector (https://channel9.msdn.com/Shows/AI-Show/Introducing-Azure-Anomaly-Detector) and Bring Anomaly Detector on-premise with containers support (https://channel9.msdn.com/Shows/AI-Show/Bring-Anomaly-Detector-on-premise-with-containers-support)). In this episode, we're moving one step closer to your production usage of the service. We're sharing with you three tips or best practices that you will want to know before you design your monitoring application architecture. Tip 1: Batch with "/entire" and streaming with "/last". Tip 2: Streaming in practice: you need to have a data pre-processor to load, cache, and prepare the data and make HTTP calls to get the anomalies from "/last" endpoint. Tip 3: Data preparation: the v1.0 API only accepts time series data formated in JSON; The data points need to distribute evenly on timeline; to gain best performance when there're seasonal patterns, use "period" and send 4 X period +1 data points in each HTTP call. Re-Visit Your Favorite Part of the Video: [00:46] Tip 1: Batch mode versus streaming mode in Anomaly Detector. [02:76] Tip 2: How to detect anomalies from streaming time series data with Anomaly Detector? [05:00] An interactive demo of streaming anomaly detection. [06:36] Python code of streaming anomaly detection with Anomaly Detector APIs. [07:39] Tip 3: How to preproce
Published 2019-08-27 · Watch on YouTube ↗
Availability and playback restrictions are controlled by the original channel.





