search

chickenbestlover / RNN-Time-series-Anomaly-Detection

RNN based Time-series Anomaly detector model implemented in Pytorch.
Apache License 2.0
1.17k stars 316 forks source link

anomaly scores for multivariate analysis #25

Open jchenWill opened 5 years ago

jchenWill commented 5 years ago

Hi I am trying to use your code for multivariate time series analysis. The anomaly scores calculated from different channels should be the same correct? I thought the anomaly scores aggregate the prediction errors from all the channels. In you ecg example, the anomaly score for the two channels is the same. However, in my own example, it is not the same. As for different channel, they have different anomaly score. Is there any reason?

shopeth commented 4 years ago

@chickenbestlover Thank you so much for sharing the code. It's great effort.

I have a similar question regarding multivariate data. It seems the training is done by jointly considering all channels (by mapping them to a latent space with a linear encoder). However, the anomaly detection is on each channel separately. Is there a reason for this? Have you tried using the predictions for all channels to calculate the anomaly score?