diff --git a/python/lsst/pipe/tasks/functors.py b/python/lsst/pipe/tasks/functors.py index 619c80f43..179e0f4b1 100644 --- a/python/lsst/pipe/tasks/functors.py +++ b/python/lsst/pipe/tasks/functors.py @@ -28,6 +28,7 @@ "ComputePixelScale", "ConvertPixelToArcseconds", "ConvertPixelSqToArcsecondsSq", "ConvertDetectorAngleToPositionAngle", + "ConvertDetectorAngleErrToPositionAngleErr", "ReferenceBand", "Photometry", "NanoJansky", "NanoJanskyErr", "LocalPhotometry", "LocalNanojansky", "LocalNanojanskyErr", "LocalDipoleMeanFlux", @@ -1394,6 +1395,46 @@ def getPositionAngleFromDetectorAngle(self, theta, cd11, cd12, cd21, cd22): # Position angle of vector from (RA1, Dec1) to (RA2, Dec2) return np.rad2deg(self.computePositionAngle(ra1, dec1, ra2, dec2)) + def getPositionAngleErrFromDetectorAngleErr(self, theta, theta_err, cd11, + cd12, cd21, cd22): + """Compute position angle error (E of N) from detector angle error. + + Parameters + ---------- + theta : `float` + detector angle [radian] + theta_err : `float` + detector angle err [radian] + cd11 : `float` + [1, 1] element of the local Wcs affine transform. + cd12 : `float` + [1, 2] element of the local Wcs affine transform. + cd21 : `float` + [2, 1] element of the local Wcs affine transform. + cd22 : `float` + [2, 2] element of the local Wcs affine transform. + + Returns + ------- + Position Angle Error: `~pandas.Series` + Position angle error in degrees + + Notes + ----- + The error is estimated by evaluating the position angle at + ``theta ± theta_err`` and taking half the absolute difference. + """ + pa_plus = self.getPositionAngleFromDetectorAngle( + theta + theta_err, cd11, cd12, cd21, cd22 + ) + pa_minus = self.getPositionAngleFromDetectorAngle( + theta - theta_err, cd11, cd12, cd21, cd22 + ) + # Wrap difference into (-180, 180] to handle the ±180 deg boundary. + delta = pa_plus - pa_minus + delta = (delta + 180) % 360 - 180 + return np.abs(delta) / 2 + class ComputePixelScale(LocalWcs): """Compute the local pixel scale from the stored CDMatrix. @@ -1554,6 +1595,53 @@ def _func(self, df): ) +class ConvertDetectorAngleErrToPositionAngleErr(LocalWcs): + """Compute a position angle error from a detector angle error and the + stored CDMatrix. + + Returns + ------- + position angle error : degrees + """ + + name = "PositionAngleErr" + + def __init__( + self, + theta_col, + theta_err_col, + colCD_1_1, + colCD_1_2, + colCD_2_1, + colCD_2_2, + **kwargs + ): + self.theta_col = theta_col + self.theta_err_col = theta_err_col + super().__init__(colCD_1_1, colCD_1_2, colCD_2_1, colCD_2_2, **kwargs) + + @property + def columns(self): + return [ + self.theta_col, + self.theta_err_col, + self.colCD_1_1, + self.colCD_1_2, + self.colCD_2_1, + self.colCD_2_2 + ] + + def _func(self, df): + return self.getPositionAngleErrFromDetectorAngleErr( + df[self.theta_col], + df[self.theta_err_col], + df[self.colCD_1_1], + df[self.colCD_1_2], + df[self.colCD_2_1], + df[self.colCD_2_2] + ) + + class ReferenceBand(Functor): """Return the band used to seed multiband forced photometry. @@ -1992,14 +2080,7 @@ def _func(self, df): class MomentsBase(Functor): - """Base class for functors that use shape moments and localWCS - - Attributes - ---------- - is_covariance : bool - Whether the shape columns are terms of a covariance matrix. If False, - they will be assumed to be terms of a correlation matrix instead. - """ + """Base class for functors that use shape moments and localWCS""" is_covariance: bool = True diff --git a/tests/test_functors.py b/tests/test_functors.py index 2abd6d906..b33883cab 100644 --- a/tests/test_functors.py +++ b/tests/test_functors.py @@ -53,8 +53,8 @@ MomentsG1Sky, MomentsG2Sky, MomentsTraceSky, CorrelationIuuSky, CorrelationIvvSky, CorrelationIuvSky, SemimajorAxisFromCorrelation, SemiminorAxisFromCorrelation, - PositionAngleFromCorrelation - ) + PositionAngleFromCorrelation, + ConvertDetectorAngleErrToPositionAngleErr) ROOT = os.path.abspath(os.path.dirname(__file__)) @@ -860,6 +860,202 @@ def testConvertPixelToArcseconds(self): atol=1e-16, rtol=1e-16)) + def testConvertDetectorAngleErrToPositionAngleErr(self): + """Test conversion of position angle err in detector degrees to + position angle err on sky. + + Requires a similar setup to testConvertDetectorAngleToPositionAngle. + """ + dipoleSep = 10 + ixx = 10 + testPixelDeltas = np.random.uniform(-100, 100, size=(self.nRecords, 2)) + + for dec in np.linspace(-80, 80, 10): + for theta in (-35, 0, 90): + for x, y in zip( + np.random.uniform(2 * 1109.99981456774, size=10), + np.random.uniform(2 * 560.018167811613, size=10)): + wcs = self._makeWcs(dec=dec, theta=theta) + cd_matrix = wcs.getCdMatrix() + + self.dataDict["dipoleSep"] = np.full(self.nRecords, + dipoleSep) + self.dataDict["ixx"] = np.full(self.nRecords, ixx) + self.dataDict["slot_Centroid_x"] = np.full(self.nRecords, + x) + self.dataDict["slot_Centroid_y"] = np.full(self.nRecords, + y) + self.dataDict["someCentroid_x"] = x + testPixelDeltas[:, 0] + self.dataDict["someCentroid_y"] = y + testPixelDeltas[:, 1] + self.dataDict["orientation"] = np.arctan2( + self.dataDict["someCentroid_y"] - self.dataDict[ + "slot_Centroid_y"], + self.dataDict["someCentroid_x"] - self.dataDict[ + "slot_Centroid_x"], + ) + self.dataDict["orientation_err"] = np.abs(np.arctan2( + self.dataDict["someCentroid_y"] - self.dataDict[ + "slot_Centroid_y"], + self.dataDict["someCentroid_x"] - self.dataDict[ + "slot_Centroid_x"], + )) * 0.001 + + self.dataDict["base_LocalWcs_CDMatrix_1_1"] = np.full( + self.nRecords, + cd_matrix[0, 0]) + self.dataDict["base_LocalWcs_CDMatrix_1_2"] = np.full( + self.nRecords, + cd_matrix[0, 1]) + self.dataDict["base_LocalWcs_CDMatrix_2_1"] = np.full( + self.nRecords, + cd_matrix[1, 0]) + self.dataDict["base_LocalWcs_CDMatrix_2_2"] = np.full( + self.nRecords, + cd_matrix[1, 1]) + df = self.getMultiIndexDataFrame(self.dataDict) + + # Test detector angle err to position angle err conversion + func = ConvertDetectorAngleErrToPositionAngleErr( + "orientation", + "orientation_err", + "base_LocalWcs_CDMatrix_1_1", + "base_LocalWcs_CDMatrix_1_2", + "base_LocalWcs_CDMatrix_2_1", + "base_LocalWcs_CDMatrix_2_2" + ) + val = self._funcVal(func, df) + + # Errors must be non-negative and finite across all + # WCS configurations, declinations, and pixel positions. + self.assertTrue(np.all(val.to_numpy() >= 0), + "PA errors must be non-negative") + self.assertTrue(np.all(np.isfinite(val.to_numpy())), + "PA errors must be finite") + + def testPositionAngleErrPureRotationWcs(self): + """PA error for a pure scaling WCS is exactly rad2deg(theta_err). + + For CD = diag(s, -s) the sky direction of a pixel unit vector at + angle theta is PA(theta) = theta + pi/2 (flat-sky), so + |dPA/dtheta| = 1 everywhere and the propagated error is simply + np.rad2deg(theta_err) regardless of theta. This gives an + independent analytical expectation that does not depend on the + implementation under test. + """ + s = 5e-5 # degrees/pixel + cd11, cd12, cd21, cd22 = s, 0.0, 0.0, -s + local_wcs = LocalWcs("cd11", "cd12", "cd21", "cd22") + + thetas = pd.Series([0.0, np.pi / 4, np.pi / 3, -np.pi / 6, 1.2, -0.8]) + theta_err = 0.01 # radians + n = len(thetas) + cd11_s = pd.Series(np.full(n, cd11)) + cd12_s = pd.Series(np.full(n, cd12)) + cd21_s = pd.Series(np.full(n, cd21)) + cd22_s = pd.Series(np.full(n, cd22)) + + pa_err = local_wcs.getPositionAngleErrFromDetectorAngleErr( + thetas, theta_err, cd11_s, cd12_s, cd21_s, cd22_s + ) + + expected_pa_err_deg = np.rad2deg(theta_err) + np.testing.assert_allclose( + pa_err.to_numpy(), expected_pa_err_deg, rtol=1e-4, atol=0 + ) + + def testPositionAngleErrWrapAround(self): + """PA error is correct when PA crosses the ±180 deg boundary. + + For CD = diag(s, -s) and theta = pi/2, PA(theta) = pi deg exactly + (the wrap boundary). With a small theta_err: + pa_plus = PA(pi/2 + theta_err) wraps to -180 + rad2deg(theta_err) + pa_minus = PA(pi/2 - theta_err) = +180 - rad2deg(theta_err) + + Naive subtraction gives ~-360 deg → apparent error ~180 deg (wrong). + The wrap-corrected implementation should return rad2deg(theta_err). + """ + s = 5e-5 # degrees/pixel + cd11, cd12, cd21, cd22 = s, 0.0, 0.0, -s + local_wcs = LocalWcs("cd11", "cd12", "cd21", "cd22") + + theta_center = np.pi / 2 # PA(theta_center) = 180 deg → wrap boundary + theta_err = 0.01 # radians + + cd_s = lambda v: pd.Series([v]) # noqa: E731 + pa_err = local_wcs.getPositionAngleErrFromDetectorAngleErr( + pd.Series([theta_center]), + theta_err, + cd_s(cd11), cd_s(cd12), cd_s(cd21), cd_s(cd22), + ) + + expected_pa_err_deg = np.rad2deg(theta_err) + # Correct result is ~0.57 deg; naive (unwrapped) result would be ~180 deg. + np.testing.assert_allclose( + pa_err.to_numpy(), expected_pa_err_deg, rtol=1e-4, atol=0 + ) + self.assertLess(float(pa_err.iloc[0]), 1.0) + + def testPositionAngleErrAccuracy(self): + """Validate PA error accuracy by comparing to a Monte Carlo distribution. + + Draw n_samples values of theta from N(theta_center, theta_err), compute + the position angle for each sample using getPositionAngleFromDetectorAngle, + and compare the circular standard deviation of the resulting PA + distribution to the functor output. + + Two CD matrices are used to test normal scaling vs shearing: + - Pure scaling (CD = diag(s, -s)): PA(theta) = theta + pi/2, Jacobian = 1. + - Sheared WCS: Jacobian varies with theta, exercising the non-linear path. + """ + rng = np.random.default_rng(42) + n_samples = 2000 + theta_err = 0.01 + tolerance = 0.05 # allow 5 % relative error + + s = 5.5e-5 # degrees/pixel. Do I need to be super accurate? + # Need to test against both?? Probably??? + cd_matrices = { + "pure_scaling": np.array([[s, 0.0], [0.0, -s]]), + "sheared": np.array([[s, 0.4 * s], [-0.2 * s, -s]]), + } + + for label, cd in cd_matrices.items(): + cd11, cd12, cd21, cd22 = cd[0, 0], cd[0, 1], cd[1, 0], cd[1, 1] + local_wcs = LocalWcs("cd11", "cd12", "cd21", "cd22") + + for theta_center in [0.0, 0.5, -1.2]: + theta_samples = rng.normal(theta_center, theta_err, size=n_samples) + n = len(theta_samples) + + pa_samples_deg = local_wcs.getPositionAngleFromDetectorAngle( + pd.Series(theta_samples), + pd.Series(np.full(n, cd11)), + pd.Series(np.full(n, cd12)), + pd.Series(np.full(n, cd21)), + pd.Series(np.full(n, cd22)), + ) + pa_rad = np.deg2rad(pa_samples_deg.to_numpy()) + # Make sure circular standard deviation handles the +/-180 deg + # wrap correctly. + R = np.abs(np.mean(np.exp(1j * pa_rad))) + mc_std_deg = np.rad2deg(np.sqrt(-2.0 * np.log(R))) + + pa_err = local_wcs.getPositionAngleErrFromDetectorAngleErr( + pd.Series([theta_center]), + theta_err, + pd.Series([cd11]), + pd.Series([cd12]), + pd.Series([cd21]), + pd.Series([cd22]), + ) + + self.assertAlmostEqual( + float(pa_err.iloc[0]), + mc_std_deg, + delta=mc_std_deg * tolerance, + msg=f"MC vs functor mismatch for {label}, theta_center={theta_center}", + ) + def _makeWcs(self, dec=53.1595451514076, theta=0): """Create a wcs from real CFHT values, rotated by an optional theta.