Repository navigation
DetectSpikeBy(Ssa|Iid) confidence param should be double #4058
Description
This one is pretty simple. Confidence parameter in DetectIidSpike & DetectSpikeBySsa should be of type double. It makes no sense that it is of type Int32 since it is being cast to double when creating SsaSpikeDetector.Options() anyway.
Same thing goes for DetectChangePointBySsa, DetectIidChangePoint and possibly others.
Right now it is not possible to search only for "extreme spikes" (confidence = ~99.9 or even more) using these extension methods.
Thanks
Esso
Metadata
Metadata
Assignees
Labels
No labels
Activity
esso23 commentedon Sep 3, 2019
I would recommend raising the priority to highest possible because this renders all the ChangePoint/Spike detection algorithms (almost) useless. We had to modify the ML.NET library and recompile it to be able to use them. We are regularly using confidences like 99.9999999 etc. 99.0 is just waaaay too low for many cases and produces unwanted outputs.
Also, there is no workaround, as these extension methods are the only way how to call these algorithms, all other methods/classes are either private or internal.
codemzs commentedon Sep 3, 2019
@esso23 looking into it now
esso23 commentedon Feb 12, 2021
Honesly I can't believe that such an important thing is still not fixed.
This really makes the entire TimeSeries library unusable in so many real-world scenarios.
TimeSeries - fix confidence parameter type for some detectors (#4058) (…
Merge from main repository (#1)
Actually merge from main (#2)
Update editor config and format samples (#5893)