search

twitter / AnomalyDetection

Anomaly Detection with R
GNU General Public License v3.0
3.55k stars 779 forks source link

[Q] How to use this algorithm in real-time/on-line? #91

Open N1h1l1sT opened 6 years ago

N1h1l1sT commented 6 years ago

So, for off-line usage, you have the dataset, you put a date/time as the 1st column in a data.frame and a numeric one as the second, and you feed it to AnomalyDetectionTs()

This way you can get an off-line graph via $plot (not being re-rendered as new data come) and the actual anomalous entries on $anoms

But what if you wish to use it not for offline data, but for streaming data? Can you get a responsive graph? and most importantly, can you get the increasing number of anomalies?

Or is the idea to merge the data into a big dataset and run it every X minutes/hours/etc with an "only_last" value?

tharejamudit commented 6 years ago

thinking of this

N1h1l1sT commented 6 years ago

Yeah, having an on-line way of getting anomalies and graph is very important..

irisava commented 5 years ago

Except for updating the data to contain the last streaming data every minutes/hours, is there any other method?