Interaction between Training Data and Sliding Window when doing Inference?

Hans Meyer 46

Hi there,

I was wondering, when doing inference on unseen data, does MVAD make a decision only based on the points in the sliding window (within the inference data), or also the data points in the training data? I.e. how is the information from the last sliding window (in the inference data) combined with training data?

romungi-MSFT 42,406 Reputation points Microsoft Employee

2021-08-17T12:46:11.507+00:00

@Hans Meyer I am not sure if this FAQ question from the documentation answers your question but the scenarios mentioned explains how the sliding window is used in streaming and batch scenario. So, it looks like in the batch scenario the inference data is used and inference in such scenarios are performed in a "moving window"
Hans Meyer 46 Reputation points

2021-08-17T12:52:45.207+00:00

The way it is formulated there makes it seem like the decision whether a point is anomalous or not is only based on the points in the sliding window. My question was then: is / in which way is the training data considered when making this decision?