Closed MuffinSpawn closed 6 years ago
Changed to using the predictive R^2 for the mapping quality metric. Added a check for low predictive R^2, triggering an map.nack in Position. Need the LT to properly test.
Added an exception for low-sample maps (i.e. when generating a map using four or fewer reflectors). The predictive R^2 penalizes for having a number of samples <= number of variables/dimensions.
Also, instead of forcing a 1 for the constant term R^2, the verification VI deliberately checks that this value is less than -500.
Also tested.
If the DSCS coordinates are not correct or there is otherwise a mismatch between measured LTCS and configured DSCS coordinates of the reference network reflectors, predictions for Mapper and DS reflector locations can be wildly inaccurate. A simple check on the accuracy of the LT->DS mapping could save a lot of time investigating why the system is not functioning properly by identifying when a bad LT->DS mapping was generated. A map.ack accompanied by an error event should suffice in terms of EMMA actions.
Although the R^2 value is computed, this value appears not to be sensitive enough for our purposes. A simple test, that lends itself well to establishing intuitive thresholds, is to convert the measured LTCS coordinates to DSCS and compare them with the configured DSCS coordinates. The discrepancies shouldn't be more than a few mm.