Fix Version/s: None
If there are masked pixels in the overscan, the MEDIAN_PER_ROW overscan correction code generates a warning:
python/lsst/ip/isr/overscan.py:344: UserWarning: Warning: converting a masked element to nan.
rowMedian = afwMath.makeStatistics(row, fitType, self.statControl).getValue()
Please remove this. I think the simplest fix is row = rowrow.mask.data in collapseArrayMedian, but there may be other occurrences. A better solution might be to stop using np.ma in favour of using stack-native code everywhere, which understands Mask planes.
Robert Lupton added a comment -
Looks good. I've no idea why my original fix stopped working, but yours is better anyway (if we have to keep np.ma)
Robert Lupton added a comment - Looks good. I've no idea why my original fix stopped working, but yours is better anyway (if we have to keep np.ma)
The fix above causes makeStatistics to raise. Switched to `numpy.ma.compressed()`, which though poorly named, does the right operation. Added unit test as well because the overscan outlier masking has apparently never been tested.