Question about completeness of benchmark/paper-benchmark.csv

[nat-hu]

Hello,

I have a quick question regarding benchmark/paper-benchmark.csv.

The reported numbers seem incomplete to me: the time series appear relatively short and contain only few anomalies, whereas I would expect more anomalies given the original series length.

Were the time series truncated or otherwise filtered for evaluation purposes, or is this file a subset of the full benchmark data?

Thanks for the clarification and for providing the benchmark.

[sarahmish]

Hi @nat-hu, can you clarify which dataset you are referring to?

In MSL and SMAP, these datasets were pre-split by the original authors into training and testing. Since the study focuses on zero-shot detection, we only applied it to the testing split.