def testObjective(self):
eps = 1E-6
psfModel = ms.FitPsfModel(ms.FitPsfControl(), 1.0,
numpy.array([0.1, -0.05, 1.0]))
obj = ms.FitProfileAlgorithm.makeObjective(self.ctrl, psfModel,
self.inputs)
resolved = geom.ellipses.Quadrupole(15.0, 12.0, 0.23)
circle = geom.ellipses.Quadrupole(1.0, 1.0, 0.0)
deltafn = geom.ellipses.Quadrupole(0.0, 0.0, 0.0)
for q in (resolved, circle, deltafn):
e = ms.MultiGaussianObjective.EllipseCore(q)
ms.MultiGaussianObjective.constrainEllipse(e, self.ctrl.minRadius,
self.ctrl.minAxisRatio)
parameters = numpy.zeros(3, dtype=float)
ms.MultiGaussianObjective.writeParameters(e, parameters)
f0 = numpy.zeros(self.inputs.getSize(), dtype=float)
obj.computeFunction(parameters, f0)
d0 = numpy.zeros((parameters.size, self.inputs.getSize()),
dtype=float).transpose()
d1 = numpy.zeros((parameters.size, self.inputs.getSize()),
dtype=float).transpose()
obj.computeDerivative(parameters, f0, d0)
for i in range(parameters.size):
parameters[i] += eps
f1a = numpy.zeros(self.inputs.getSize(), dtype=float)
obj.computeFunction(parameters, f1a)
parameters[i] -= 2.0 * eps
f1b = numpy.zeros(self.inputs.getSize(), dtype=float)
obj.computeFunction(parameters, f1b)
parameters[i] -= eps
d1[:, i] = (f1a - f1b) / (2.0 * eps)
self.assertClose(d0, d1, atol=1E-4, rtol=1E-10)
print d0
def testConvolvedModel(self):
psfModel = ms.FitPsfModel(ms.FitPsfControl(), 1.0,
numpy.array([0.1, -0.05, 1.0]))
psfMultiGaussian = psfModel.getMultiGaussian()
multiGaussian = ms.MultiGaussianRegistry.lookup(self.ctrl.profile)
parameters = numpy.array([[0.0, 0.0, 0.0], [0.0, 0.0, -16],
[0.2, -0.8, 2.3]])
sImage = lsst.afw.image.ImageD(self.bbox)
builders = []
for p in psfMultiGaussian:
psfEllipse = psfModel.ellipse.clone()
psfEllipse.scale(p.radius)
builders.extend(
ms.GaussianModelBuilder(self.x, self.y, c.flux, c.radius,
psfEllipse, p.flux)
for c in multiGaussian)
psfShapelets = psfModel.asMultiShapelet()
for row in parameters:
model = ms.FitProfileModel(self.ctrl, 1.0, row)
msf = model.asMultiShapelet(self.center).convolve(psfShapelets)
z0 = sImage.getArray()
z0[:, :] = 0.0
msf.evaluate().addToImage(sImage)
z1 = numpy.zeros(z0.shape, dtype=float)
for b in builders:
b.update(ms.MultiGaussianObjective.readParameters(row))
z1 += b.getModel().reshape(*z1.shape)
self.assertClose(z0, z1)
self.assert_(numpy.isfinite(z0).all())
self.assert_(numpy.isfinite(z1).all())
def testObjective(self):
eps = 1E-6
ctrl = ms.FitPsfControl()
image = lsst.afw.image.ImageD(5, 5)
nParameters = 3
nData = image.getBBox().getArea()
nTests = 10
center = geom.Point2D(2.0, 2.0)
xGrid, yGrid = numpy.meshgrid(numpy.arange(-2, 3), numpy.arange(-2, 3))
image.getArray()[:, :] = 1.0 * numpy.exp(-0.5 * (xGrid**2 + yGrid**2))
image.getArray()[:, :] += numpy.random.randn(5, 5) * 0.1
inputs = ms.ModelInputHandler(image, center,
image.getBBox(lsst.afw.image.PARENT))
obj = ms.FitPsfAlgorithm.makeObjective(ctrl, inputs)
parameters = numpy.random.rand(nTests, nParameters) * 0.5
for i in range(nTests):
f0 = numpy.zeros(nData, dtype=float)
obj.computeFunction(parameters[i, :], f0)
f1 = obj.getModel() * obj.getAmplitude() - inputs.getData()
model = ms.FitPsfModel(ctrl, obj.getAmplitude(), parameters[i, :])
self.assertClose(
model.outer[0],
ctrl.peakRatio * model.inner[0] * ctrl.radiusRatio**2)
self.assertEqual(model.radiusRatio, ctrl.radiusRatio)
image2 = lsst.afw.image.ImageD(5, 5)
multiShapeletFunc = model.asMultiShapelet(center)
multiShapeletFunc.evaluate().addToImage(image2)
f2 = (image2.getArray().ravel() - inputs.getData())
multiGaussian = model.getMultiGaussian()
builder1 = ms.GaussianModelBuilder(inputs.getX(), inputs.getY(),
multiGaussian[0].flux,
multiGaussian[0].radius)
builder2 = ms.GaussianModelBuilder(inputs.getX(), inputs.getY(),
multiGaussian[1].flux,
multiGaussian[1].radius)
builder1.update(model.ellipse)
builder2.update(model.ellipse)
f3 = builder1.getModel() + builder2.getModel() - inputs.getData()
self.assertClose(f0, f1)
self.assertClose(f0, f2)
self.assertClose(f0, f3)
d0 = numpy.zeros((nParameters, nData), dtype=float).transpose()
d1 = numpy.zeros((nParameters, nData), dtype=float).transpose()
obj.computeDerivative(parameters[i, :], f0, d0)
for j in range(nParameters):
parameters[i, j] += eps
f1a = numpy.zeros(nData, dtype=float)
obj.computeFunction(parameters[i, :], f1a)
parameters[i, j] -= 2.0 * eps
f1b = numpy.zeros(nData, dtype=float)
obj.computeFunction(parameters[i, :], f1b)
d1[:, j] = (f1a - f1b) / (2.0 * eps)
parameters[i, j] += eps
self.assertClose(d0, d1, rtol=1E-10, atol=1E-8)
def __init__(self, source, exposure, name="exp", psfCtrl=None, ctrl=None):
if ctrl is None:
if name == "exp":
config = ms.FitExponentialConfig()
else:
config = ms.FitDeVaucouleurConfig()
ctrl = config.makeControl()
if psfCtrl is None:
psfCtrl = ms.FitPsfControl()
self.ctrl = ctrl
self.ctrl.name = "multishapelet." + name
self.psfModel = ms.FitPsfModel(psfCtrl, source)
self.saved = self.Model(self.ctrl, source)
self.center = source.getCentroid()
self.shape = source.getShape()
self.inputs = self.Algorithm.adjustInputs(self.ctrl, self.psfModel,
self.shape,
source.getFootprint(),
exposure, self.center)
self.footprint = self.inputs.getFootprint()
self.bbox = self.footprint.getBBox()
self.image = lsst.afw.image.ImageD(self.bbox)
self.image.getArray()[:, :] = exposure.getMaskedImage().getImage(
).getArray()[self.bbox.getSlices()]
initialEllipse = ms.MultiGaussianObjective.EllipseCore(self.shape)
opt = self.Algorithm.makeOptimizer(self.ctrl, self.psfModel,
initialEllipse, self.inputs)
maxIter = opt.getControl().maxIter
self.iterations = [self.Iteration(opt, self)]
for self.iterCount in range(maxIter):
opt.step()
self.iterations.append(self.Iteration(opt, self))
if opt.getState() & ms.HybridOptimizer.FINISHED:
break
self.model = self.Model(self.iterations[-1].model)
self.Algorithm.fitShapeletTerms(self.ctrl, self.psfModel, self.inputs,
self.model)