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

[BUG]: Computing normal stats is wrong #36

Closed dagap closed 4 years ago

dagap commented 4 years ago

The following code:

organized[t].append(predictions[step+t-args.prediction_window_size][args.prediction_window_size-1-step])

Should this not be simply:

organized[t].append(predictions[step+t-args.prediction_window_size][step])

Otherwise, I think it is reversing the vector.

Also, I think you should use the absolute value for computing the error vector.