03-09-2017 03:00 AM
I have just had a system go down because I had some complex data that had more than one data point at re=0.
The linear fit vi returned NaN for slope and intercept.
The function then returned a nonsense error code that seems totally unrelated to the problem.
Error -20055 occurred at an unidentified location
Possible reason(s):
Analysis: The number of categories or samples must be greater than one.
The data was totally valid and the fitting should have been trivial. The implementation is hidden in a DLL, so I can't see why it failed.
Is there something obvious I am missing here?
Snippet to reproduce attached.
Solved! Go to Solution.
03-09-2017 03:30 AM - edited 03-09-2017 03:31 AM
03-09-2017 02:45 PM
Bug was there in 2011.
If there are 2 points with x=0, least absolute residual (and only it) fails.
03-09-2017 03:17 PM - edited 03-09-2017 03:18 PM
Yes, definitely a bug. Substituting machine epsilon (or even 1e-200!) for zero works fine. 😮
(It is annoying that the code is trapped inside a dll, else we could see what the problem is. :()
Is there a CAR?
03-09-2017 03:33 PM
@altenbach wrote:
Yes, definitely a bug. Substituting machine epsilon (or even 1e-200!) for zero works fine. 😮
(It is annoying that the code is trapped inside a dll, else we could see what the problem is. :()
Is there a CAR?
Hopefully they will CAR it if its not already. Anyone at NI confirm if one exists for this?
Its easy enough to work around when you know the problem exists.
Not so easy when it crashes a production system on the other side of the world.
To say the last few days were stressful is a bit of an understatement.
03-10-2017 11:21 AM
A CAR has been created to cover this problem.
-Jim
03-10-2017 01:54 PM
Cheers Jim.