Will the anomaly detector provide stateful version?

Question

I have tried the new anomaly detector for some days, it seems workable in most of the cases. But it seems it is total stateless and I have to push all the history data for one single detection. It is very difficult to scale in our current system. So does anyone could answer when it will be a stateful version? In that case I can post only the new points. Another thing is, is the custom tuning model in the roadmap? I mean if I provide some feedback for my time series such as false positive, will it be retrained to provide better result in the future?

Answer 1

Thanks for using anomaly detector. Glad it works for your case. Actually we're planning the next version of anomaly detector. In this version, you just push the data each time it generated and the system will store these data. These data will be used to detect anomaly of incoming data points. In this new version, you can only post new points to get corresponding detection results. For your second question, about customized detection model, we're planning a way to collect user's feedback and retrain the detection model with user's feedback. Better results can be expected with this feedback loop. Hope these answers help you.