Thanks for sharing. Can you elaborate on how you grouped your timeseries into similar timeseries? What was the overhead of that?