DM-38777: LinearizeSpline linearity corrections do not anchor the spline at zero flux #258
Conversation
python/lsst/ip/isr/linearize.py
Outdated
| @@ -753,6 +756,14 @@ def __call__(self, image, **kwargs): | |||
| """ | |||
| splineCoeff = kwargs['coeffs'] | |||
| centers, values = np.split(splineCoeff, 2) | |||
| if centers[0] != 0.0 or values[0] != 0: | |||
There was a problem hiding this comment.
Is this deliberately or rather than a nested if as in cp_pipe?
There was a problem hiding this comment.
I thought it was correct as there are two issues to correct, and then discovered that giving the interpolation code centers like [0.0, 0.0, 1.0, 2.0, 3.0, 4.0, 5.0] causes an error even if the values for the two 0.0 points is identical. I've split the two issues into separate checks, which should prevent that interpolation error.
| @@ -125,8 +125,8 @@ def setUp(self): | |||
| [[1e-5, 1e-7], [1.1e-6, 1e-7], [2.1e-6, 1e-7]]], dtype=float) | |||
|
|
|||
| # Spline coefficients: should match a 1e-6 Squared solution | |||
| self.splineCoeffs = np.array([-100, 0.0, 1000, 2000, 3000, 4000, 5000, | |||
| 0.0, 0.0, 1.0, 4.0, 9.0, 16.0, 25.0]) | |||
There was a problem hiding this comment.
Hmm, presumably this would have generated at least slightly negative values between -100 and 0 before, no? Did that cause any issues? But either way, starting it at zero definitely seems to make more sense.
There was a problem hiding this comment.
I think it was just poor thinking on my part when I wrote the test. If that does match the 1e-6 squared solution, then that negative point gains flux from a correction entirely derived from the positive data, which doesn't make any sense at all.
python/lsst/ip/isr/linearize.py
Outdated
| if self._detectorSerial != detector.getSerial(): | ||
| raise RuntimeError("Detector serial numbers don't match: %s != %s" % | ||
| (self._detectorSerial, detector.getSerial())) | ||
| # DM-38778: This check fails on LATISS due to an error in the |
There was a problem hiding this comment.
Consider prefacing this and below note with TODO: Uncomment when DM-38778 is complete..., if you find TODO tags helpful.
Fix spline linearity-induced amplifier offsets.