There are many algorithms which not just give a binary classification into cloud/non-cloud valid/invalid, but probabilities of observation quality.
During monitoring, it can be valuable if the residuals can be weighted based on such a classification. This only really makes sense for continuous (auto correlated) monitoring like CUSUM, MOSUM, EWMA since for CCDC and IQR a low quality observation could inhibit sequential exceedances of the boundary.
I would propose an additional parameter for .monitor() called weights. Which would be an array with the same size as array. These weights would then be applied to the residuals, decreasing the importance of low quality observations.
Optionally another parameter scale could be introducted which would be used to automatically scale the data from 0-1. It would be a tuple of minimum and maximum value. For example (0, 100) so that raw data can be passed in without any necessary computations on the users side.
@loicdtx do you have any input on this?
There are many algorithms which not just give a binary classification into cloud/non-cloud valid/invalid, but probabilities of observation quality.
During monitoring, it can be valuable if the residuals can be weighted based on such a classification. This only really makes sense for continuous (auto correlated) monitoring like CUSUM, MOSUM, EWMA since for CCDC and IQR a low quality observation could inhibit sequential exceedances of the boundary.
I would propose an additional parameter for
.monitor()called weights. Which would be an array with the same size asarray. These weights would then be applied to the residuals, decreasing the importance of low quality observations.Optionally another parameter
scalecould be introducted which would be used to automatically scale the data from 0-1. It would be a tuple of minimum and maximum value. For example(0, 100)so that raw data can be passed in without any necessary computations on the users side.@loicdtx do you have any input on this?