Fix CGGTTS, reset the tracker after it is released.

[vicalloy]

If tracker.fit fails, tracker.reset() will not be called, and the samples will be kept for the next period.
This commit resets the tracker after tacker release.

[gwbres]

my big question in CGGTTS track fitting process, is whether we should tolerate data gaps or not.
I presume we should, because it is unrealistic to hope for 15' steady observations.
And after all, that is the whole point of a fitting algorithm

[vicalloy]

my big question in CGGTTS track fitting process, is whether we should tolerate data gaps or not. I presume we should, because it is unrealistic to hope for 15' steady observations. And after all, that is the whole point of a fitting algorithm

Yes, I have the same question. In my scenario, my sample interval is 1 second, and missing one sample is not a significant issue for the result.
Maybe we can add a ratio parameter for the minimum sample count.

[gwbres]

The tracker should simply have a gap Duration tolerance, which seems the best parameter to tune by the user (depending on requirements and application). I would also argue that the tracker's default behavior should be very tolerant (tolerate almost "any" gap) for very basic applications. To me that would be 3 points per CvPeriod. Which is incompatible with high quality CGGTTS, but just works. In your 1Hz (""fast"" observation application), you could for example specify a 30s tolerance, which would be compatible with high quality CGGTTTS