def testKeplerSplineError(self):
# Big gap.
time = np.concatenate([np.arange(0, 1, 0.1), [2]])
flux = np.sin(time)
with self.assertRaises(kepler_spline.SplineError):
kepler_spline.kepler_spline(time, flux, bkspace=0.5)
def testKeplerSplineSine(self):
# Fit a sine wave.
time = np.arange(0, 10, 0.1)
flux = np.sin(time)
# Expect very close fit with no outliers removed.
spline, mask = kepler_spline.kepler_spline(time, flux, bkspace=0.5)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 1e-4)
self.assertTrue(np.all(mask))
# Add some outliers.
flux[35] = 10
flux[77] = -3
flux[95] = 2.9
# Expect a close fit with outliers removed.
spline, mask = kepler_spline.kepler_spline(time, flux, bkspace=0.5)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 1e-4)
self.assertEqual(np.sum(mask), 97)
self.assertFalse(mask[35])
self.assertFalse(mask[77])
self.assertFalse(mask[95])
# Increase breakpoint spacing. Fit is not quite as close.
spline, mask = kepler_spline.kepler_spline(time, flux, bkspace=1)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 2e-3)
self.assertEqual(np.sum(mask), 97)
self.assertFalse(mask[35])
self.assertFalse(mask[77])
self.assertFalse(mask[95])
def testFitSine(self):
# Fit a sine wave.
time = np.arange(0, 10, 0.1)
flux = np.sin(time)
# Expect very close fit with no outliers removed.
spline, mask = kepler_spline.kepler_spline(time, flux, bkspace=0.5)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 1e-4)
self.assertTrue(np.all(mask))
# Add some outliers.
flux[35] = 10
flux[77] = -3
flux[95] = 2.9
# Expect a close fit with outliers removed.
spline, mask = kepler_spline.kepler_spline(time, flux, bkspace=0.5)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 1e-4)
self.assertEqual(np.sum(mask), 97)
self.assertFalse(mask[35])
self.assertFalse(mask[77])
self.assertFalse(mask[95])
# Increase breakpoint spacing. Fit is not quite as close.
spline, mask = kepler_spline.kepler_spline(time, flux, bkspace=1)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 2e-3)
self.assertEqual(np.sum(mask), 97)
self.assertFalse(mask[35])
self.assertFalse(mask[77])
self.assertFalse(mask[95])
def testKeplerSplineError(self):
# Big gap.
time = np.concatenate([np.arange(0, 1, 0.1), [2]])
flux = np.sin(time)
with self.assertRaises(kepler_spline.SplineError):
kepler_spline.kepler_spline(time, flux, bkspace=0.5)
def testInsufficientPointsError(self):
# Empty light curve.
time = np.array([])
flux = np.array([])
with self.assertRaises(kepler_spline.InsufficientPointsError):
kepler_spline.kepler_spline(time, flux, bkspace=0.5)
# Only 3 points.
time = np.array([0.1, 0.2, 0.3])
flux = np.sin(time)
with self.assertRaises(kepler_spline.InsufficientPointsError):
kepler_spline.kepler_spline(time, flux, bkspace=0.5)
def testInsufficientPointsError(self):
# Empty light curve.
time = np.array([])
flux = np.array([])
with self.assertRaises(kepler_spline.InsufficientPointsError):
kepler_spline.kepler_spline(time, flux, bkspace=0.5)
# Only 3 points.
time = np.array([0.1, 0.2, 0.3])
flux = np.sin(time)
with self.assertRaises(kepler_spline.InsufficientPointsError):
kepler_spline.kepler_spline(time, flux, bkspace=0.5)
def testKeplerSplineCubic(self):
# Fit a cubic polynomial.
time = np.arange(0, 10, 0.1)
flux = (time - 5)**3 + 2 * (time - 5)**2 + 10
# Expect very close fit with no outliers removed. We choose maxiter=1,
# because a cubic spline will fit a cubic polynomial ~exactly, so the
# standard deviation of residuals will be ~0, which will cause some closely
# fit points to be rejected.
spline, mask = kepler_spline.kepler_spline(
time, flux, bkspace=0.5, maxiter=1)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 1e-12)
self.assertTrue(np.all(mask))
def testFitCubic(self):
# Fit a cubic polynomial.
time = np.arange(0, 10, 0.1)
flux = (time - 5)**3 + 2 * (time - 5)**2 + 10
# Expect very close fit with no outliers removed. We choose maxiter=1,
# because a cubic spline will fit a cubic polynomial ~exactly, so the
# standard deviation of residuals will be ~0, which will cause some closely
# fit points to be rejected.
spline, mask = kepler_spline.kepler_spline(
time, flux, bkspace=0.5, maxiter=1)
rmse = np.sqrt(np.mean((flux[mask] - spline[mask])**2))
self.assertLess(rmse, 1e-12)
self.assertTrue(np.all(mask))