search

khundman / telemanom

A framework for using LSTMs to detect anomalies in multivariate time series data. Includes spacecraft anomaly data and experiments from the Mars Science Laboratory and SMAP missions.
https://arxiv.org/abs/1802.04431
Other
1.02k stars 248 forks source link

Pre-scaling of train set #84

Closed tubemerc closed 2 months ago

tubemerc commented 4 months ago

As described in the README, data such as S-1 and D-1 are clearly pre-scaled according to the min/max in the test set. But data such as E-3 and E-8 have different value ranges in the training set and in normal sequences of the test set.

Is this difference in value range caused by a change of telemetry behavior in the test set, or by the fact that the training set was pre-scaled according to the min/max in the training set?

tubemerc commented 2 months ago

This issue was already mentioned in https://github.com/khundman/telemanom/issues/23.