Closed chrishylanduk closed 5 months ago
Hi @chrishylanduk
Thank you very much for raising these requirements / ideas and feature requests, this is extremely valuable and I have been taking these into accounts carefully.
In TrackMe 2.0.89, the Outliers framework has therefore been extended with the following new options (all options can / have to be set at the system level but can also be modified at the model level)
System level:
Models update level:
Adding new model:
As far as I am concerned, these new capabilities should reflect your thinking and address these properly ;-)
Any further comment, please do not hesitate, if not the next release these would be taken into account for a for further release.
This looks brilliant, thanks a lot @guilhemmarchand - looking forward to trying it out!
Hi @chrishylanduk
TrackMe 2.0.89 is live in Splunk Base - any new requirements or anything I've missed, please let me know and we'll address these in the next release ;-)
Guilhem
Is your feature request related to a problem? Please describe. Currently TrackMe uses the MLTK DensityFunction algorithm for outlier detection. This works well in some instances, but in others it does not. Changes that users might want to make, and which aren't currently possible, include:
Describe the solution you'd like TrackMe trying to meet all the above use cases, and more, doesn't seem realistic. However, if TrackMe allowed using a custom MLTK algorithm, rather than DensityFunction, then users could build exactly the models that meet their requirements. I presume there would need to be a specification of what inputs / outputs the algorithm would need to accept / return, so TrackMe could interpret them correctly.
Describe alternatives you've considered