search

linkedin / luminol

Anomaly Detection and Correlation library
Apache License 2.0
1.18k stars 215 forks source link

Calculating anomaly score for multivariate data set. #33

Open Lingaselvan opened 6 years ago

Lingaselvan commented 6 years ago

I have been using Luminol to calculate anomaly scores for a univariate data sets(Timestamp & Value) and getting good results. Now, I want to move into multivariate data sets(Timestamp & Value 1 & Value 2 & .... & Value N) and detect a single anomaly score based upon all values. I finding hard on how to proceed with this problem statement. Is there a way I can apply Luminol to this problem or could you suggest me a way on how to proceed?

Thank you.

lionsq commented 6 years ago

I don't think Luminol support multivariate. I have checked the data structure of TimeSeries supported by this module.

In the initialization code of the TimeSeries class, it is defined as follows.
def init(self, series): self.timestamps = [] self.values = [] ..... self.values.append(float(series[ts])) clearly the values is only an array with uni-variate time series