search

DeepanshuKarnwal / AD

Other
0 stars 3 forks source link

questions about anomaly labels in ue.csv #1

Open 1496nuonuo opened 10 months ago

1496nuonuo commented 10 months ago

Hi! You do very meaningful work in Oran AD. I have read many releases of the Oran AD Xapps. What confuse me most are the criterion of the anomalous labels. What does it mean? How can we judge it an anomaly datapoint or not?

DeepanshuKarnwal commented 10 months ago

Thank you. In AD xApp, if there is sudden drop in throughput or radio parameters (KPIs) of any user, those samples are labelled as anomalous otherwise normal.

1496nuonuo commented 10 months ago

So are the labels just given by observation? Or are there other more concrete formulas?

1496nuonuo commented 10 months ago

I think the mathematical formulations or some quantization of the drop in the throughput or radio parameters in a time series are a little non-trivial.

DeepanshuKarnwal commented 10 months ago

You are right! But we expect for a user to have smooth QoE i.e throughput to be maintained (normal behavior). Sudden drops (not gradual) of 70% or more in throughput kpi or similar loss in radio parameters qualify for a sample to be anomalous.

1496nuonuo commented 8 months ago

Hi! I find target throughput in the metrics collection ''ue.csv'' in the Oran-sc ad xapps. Would you tell me how to compute the target throughput in this collection? Thanks!