feat: add ~variance~ standard deviation selector

[michaelpun]

This PR adds VarianceSelector class which straightforwardly returns the variance of a metric.

From the Selector base class, I think this is within the spirit of the class to return an aggregation of the metric but due to the name "Selector" I wonder if this goes against the spirit due to it no longer having the same units as the original metric.

[padix-key]

Thanks for the feature. This is a more exotic use case, but I still think this can be useful for a broader audience and probably an implementation as Selector makes the most sense.

However, I also share your concern that we are not actually 'selecting' a value. A possible solution to this would be to change the name to VarianceAggregation or something similar.
To make the VarianceSelector use the same units as its input, one could also think about returning the standard deviation instead of the variance, if this still fulfills the purpose.

[michaelpun]

Thanks for the feedback!

To make the VarianceSelector use the same units as its input, one could also think about returning the standard deviation instead of the variance, if this still fulfills the purpose.

I think this is a great suggestion. After using the VarianceSelector for a bit, I think having the same units would be good both for the code and for the ease of interpretability on the user's end. I've replaced VarianceSelector with StandardDeviationSelector in 38e6b4f. Let me know if the name is too long.

However, I also share your concern that we are not actually 'selecting' a value. A possible solution to this would be to change the name to VarianceAggregation or something similar.

I think this could make sense. If I were to rename the now-called StandardDeviationSelector to StandardDeviationAggregator would it make sense to also rename MeanSelector since the value it produces is not a member of the set of outputted values? Or would the "selector" distinction just be for metrics that are directly tied to the random variable and "aggregator" just be for higher order statistics and other aggregations?

[padix-key]

Let me know if the name is too long.

Would DeviationAggregator also suffice?

Or would the "selector" distinction just be for metrics that are directly tied to the random variable and "aggregator" just be for higher order statistics and other aggregations?

I would say so. But maybe a third opinion would be good here. @danielkovtun @emalgorithm @dtischer Do you prefer StandardDeviationAggregator, DeviationAggregator, StandardDeviationSelector or DeviationSelector?