Statistical Process Control in Python

Fun memories.

We have successfully replaced thousands of complicated deep net time series based anomaly detectors at a FANG with statistical (non parametric) process control ones.

They use 3 to 4 orders lower number of trained parameters and has just enough complexity that a team of 3 or four can handle several thousands of such streams.

The amount of baby sitting that deep net models needed was astronomical, debugging and understanding what has happened quite opaque.

For small teams, with limited resources I would still heavily recommend stats based models for time series anomaly detection.

May not be your best career move now for political reasons. Those making massive bets do not like to confront that some of their bets might have not been well placed.