Support for make-series, series_fir, series_subtract

[lothar7]

If would be nice if we could generate moving averages using kusto-loco as described here
https://www.aizoo.info/post/kusto-lightning-fact-6-series-fir-generating-moving-averages-for-time-series-data

Might be other ways of doing it though that perhaps doesn't need all these functions

[NeilMacMullen]

make_series etc offer a huge amount of flexibillity but look like they are correspondingly difficult to implement. If all you want to do is effectively run a function over a sliding window and extend the result into an additional column then it could be done with a custom function that contains an accumulator.

The main problem with this approach is that there's currently no signal to reset the function implementation so you'd have to create a new context before each query or perform the reset manually. It's also a lot less flexible - performing calculations over grouped sets would required special logic inside the function

[lothar7]

In my specific usecase i need to calculate a moving average for the last x days as a timeseries so i am looking for ways of solving it. If it has to be a custom function then so be it. (Althoufh it would be nice if it was possible using just plain kusto). Can you elaborate on the special logic you would want for grouped sets?

[NeilMacMullen]

The CustomFunctions sample code contains this example method.

So a simple variation like

[KustoImplementation(Keyword = "rolledAverage")]
public partial class RolledAverageFunction
{
   private Dictionary<string,CircularBuffer> groupWindows. = new....
    public static string Impl(string groupid,long n)
    {
      var window = groupWindows.GetOrAdd(groupId);
     window.Add(n);
     return window.Average();
    }
}

would kind of work as a hack. As I say, the main issue is that there is currently no way to reset the window buffers between queries (unless you fetch them from a static provider and poke that from outside the query engine).