Coverage for tests/test_kernelPca.py : 12%

import unittest 

import lsst.utils.tests 

import lsst.afw.image as afwImage 

import lsst.afw.math as afwMath 

import lsst.afw.geom as afwGeom 

import lsst.ip.diffim as ipDiffim 

import lsst.ip.diffim.diffimTools as diffimTools 

import lsst.log.utils as logUtils 

import lsst.pex.config as pexConfig 

logUtils.traceSetAt("ip.diffim", 4) 

class DiffimTestCases(lsst.utils.tests.TestCase): 

def setUp(self): 

self.config = ipDiffim.ImagePsfMatchTask.ConfigClass() 

self.config.kernel.name = "DF" 

self.subconfig = self.config.kernel.active 

self.kList = ipDiffim.makeKernelBasisList(self.subconfig) 

self.policy = pexConfig.makePolicy(self.subconfig) 

self.policy.set("useRegularization", False) 

def tearDown(self): 

del self.config 

del self.policy 

del self.kList 

def makeCandidate(self, kSum, x, y, size=51): 

mi1 = afwImage.MaskedImageF(afwGeom.Extent2I(size, size)) 

mi1.getVariance().set(1.0) # avoid NaNs 

mi1[size//2, size//2, afwImage.LOCAL] = (1, 0x0, 1) 

mi2 = afwImage.MaskedImageF(afwGeom.Extent2I(size, size)) 

mi2.getVariance().set(1.0) # avoid NaNs 

mi2[size//2, size//2, afwImage.LOCAL] = (kSum, 0x0, kSum) 

kc = ipDiffim.makeKernelCandidate(x, y, mi1, mi2, self.policy) 

return kc 

def testGaussian(self, size=51): 

gaussFunction = afwMath.GaussianFunction2D(2, 3) 

gaussKernel = afwMath.AnalyticKernel(size, size, gaussFunction) 

imagePca1 = ipDiffim.KernelPcaD() # mean subtract 

imagePca2 = ipDiffim.KernelPcaD() # don't mean subtract 

kpv1 = ipDiffim.KernelPcaVisitorF(imagePca1) 

kpv2 = ipDiffim.KernelPcaVisitorF(imagePca2) 

kRefIm = None 

for i in range(100): 

kImage1 = afwImage.ImageD(gaussKernel.getDimensions()) 

gaussKernel.computeImage(kImage1, False) 

kImage1 *= 10000 # to get some decent peak source counts 

kImage1 += 10 # to get some sky background noise 

if kRefIm is None: 

kRefIm = kImage1 

kImage1 = diffimTools.makePoissonNoiseImage(kImage1) 

kImage2 = afwImage.ImageD(kImage1, True) 

imagePca1.addImage(kImage1, 1.0) 

imagePca2.addImage(kImage2, 1.0) 

kpv1.subtractMean() 

imagePca1.analyze() 

imagePca2.analyze() 

pcaBasisList1 = kpv1.getEigenKernels() 

pcaBasisList2 = kpv2.getEigenKernels() 

eVal1 = imagePca1.getEigenValues() 

eVal2 = imagePca2.getEigenValues() 

# First term is far more signficant without mean subtraction 

self.assertGreater(eVal2[0], eVal1[0]) 

# Last term basically zero with mean subtraction 

self.assertAlmostEqual(eVal1[-1], 0.0) 

# Extra image with mean subtraction 

self.assertEqual(len(pcaBasisList1), (len(eVal1) + 1)) 

# Same shape 

self.assertEqual(len(pcaBasisList2), len(eVal2)) 

# Mean kernel close to kRefIm 

kImageM = afwImage.ImageD(gaussKernel.getDimensions()) 

pcaBasisList1[0].computeImage(kImageM, False) 

for y in range(kRefIm.getHeight()): 

for x in range(kRefIm.getWidth()): 

self.assertLess(abs(kRefIm[x, y, afwImage.LOCAL] - kImageM[x, y, afwImage.LOCAL]) / 

kRefIm[x, y, afwImage.LOCAL], 0.2) 

# First mean-unsubtracted Pca kernel close to kRefIm (normalized to peak of 1.0) 

kImage0 = afwImage.ImageD(gaussKernel.getDimensions()) 

pcaBasisList2[0].computeImage(kImage0, False) 

maxVal = afwMath.makeStatistics(kRefIm, afwMath.MAX).getValue(afwMath.MAX) 

kRefIm /= maxVal 

for y in range(kRefIm.getHeight()): 

for x in range(kRefIm.getWidth()): 

self.assertLess(abs(kRefIm[x, y, afwImage.LOCAL] - kImage0[x, y, afwImage.LOCAL]) / 

kRefIm[x, y, afwImage.LOCAL], 0.2) 

def testImagePca(self): 

# Test out the ImagePca behavior 

kc1 = self.makeCandidate(1, 0.0, 0.0) 

kc1.build(self.kList) 

kc2 = self.makeCandidate(2, 0.0, 0.0) 

kc2.build(self.kList) 

kc3 = self.makeCandidate(3, 0.0, 0.0) 

kc3.build(self.kList) 

imagePca = ipDiffim.KernelPcaD() 

kpv = ipDiffim.KernelPcaVisitorF(imagePca) 

kpv.processCandidate(kc1) 

kpv.processCandidate(kc2) 

kpv.processCandidate(kc3) 

imagePca.analyze() 

eigenImages = imagePca.getEigenImages() 

# NOTE : this needs to be changed once ticket #1649 is resolved 

for i in range(len(eigenImages)): 

for j in range(i, len(eigenImages)): 

print(i, j, afwImage.innerProduct(eigenImages[i], eigenImages[j])) 

def testEigenValues(self): 

kc1 = self.makeCandidate(1, 0.0, 0.0) 

kc1.build(self.kList) 

kc2 = self.makeCandidate(2, 0.0, 0.0) 

kc2.build(self.kList) 

kc3 = self.makeCandidate(3, 0.0, 0.0) 

kc3.build(self.kList) 

imagePca = ipDiffim.KernelPcaD() 

kpv = ipDiffim.KernelPcaVisitorF(imagePca) 

kpv.processCandidate(kc1) 

kpv.processCandidate(kc2) 

kpv.processCandidate(kc3) 

imagePca.analyze() 

eigenImages = imagePca.getEigenImages() 

eigenValues = imagePca.getEigenValues() 

# took in 3 images 

self.assertEqual(len(eigenImages), 3) 

self.assertEqual(len(eigenValues), 3) 

# all the same shape, only 1 eigenvalue 

self.assertAlmostEqual(eigenValues[0], 1.0) 

self.assertAlmostEqual(eigenValues[1], 0.0) 

self.assertAlmostEqual(eigenValues[2], 0.0) 

def testMeanSubtraction(self): 

kc1 = self.makeCandidate(1, 0.0, 0.0) 

kc1.build(self.kList) 

kc2 = self.makeCandidate(2, 0.0, 0.0) 

kc2.build(self.kList) 

kc3 = self.makeCandidate(3, 0.0, 0.0) 

kc3.build(self.kList) 

imagePca = ipDiffim.KernelPcaD() 

kpv = ipDiffim.KernelPcaVisitorF(imagePca) 

kpv.processCandidate(kc1) 

kpv.processCandidate(kc2) 

kpv.processCandidate(kc3) 

kpv.subtractMean() # subtract it *from* imagePca 

imagePca.analyze() 

eigenImages = imagePca.getEigenImages() 

eigenValues = imagePca.getEigenValues() 

# took in 3 images 

self.assertEqual(len(eigenImages), 3) 

self.assertEqual(len(eigenValues), 3) 

# all the same shape, mean subtracted, so *no* eigenvalues 

self.assertAlmostEqual(eigenValues[0], 0.0) 

self.assertAlmostEqual(eigenValues[1], 0.0) 

self.assertAlmostEqual(eigenValues[2], 0.0) 

# finally, since imagePca normalizes by the sum, this should 

# have central pixel value 1.0 and the rest 0.0 

imageMean = kpv.returnMean() 

rows = imageMean.getHeight() 

cols = imageMean.getWidth() 

for y in range(rows): 

for x in range(cols): 

if x == cols // 2 and y == rows // 2: 

self.assertAlmostEqual(imageMean[x, y, afwImage.LOCAL], 1.0) 

else: 

self.assertAlmostEqual(imageMean[x, y, afwImage.LOCAL], 0.0) 

def testVisit(self, nCell=3): 

imagePca = ipDiffim.KernelPcaD() 

kpv = ipDiffim.makeKernelPcaVisitor(imagePca) 

sizeCellX = self.policy.get("sizeCellX") 

sizeCellY = self.policy.get("sizeCellY") 

kernelCellSet = afwMath.SpatialCellSet(afwGeom.Box2I(afwGeom.Point2I(0, 

0), 

afwGeom.Extent2I(sizeCellX * nCell, 

sizeCellY * nCell)), 

sizeCellX, 

sizeCellY) 

for candX in range(nCell): 

for candY in range(nCell): 

if candX == nCell // 2 and candY == nCell // 2: 

kc = self.makeCandidate(100.0, 

candX * sizeCellX + sizeCellX // 2, 

candY * sizeCellY + sizeCellY // 2) 

else: 

kc = self.makeCandidate(1.0, 

candX * sizeCellX + sizeCellX // 2, 

candY * sizeCellY + sizeCellY // 2) 

kc.build(self.kList) 

kernelCellSet.insertCandidate(kc) 

kernelCellSet.visitCandidates(kpv, 1) 

imagePca.analyze() 

eigenImages = imagePca.getEigenImages() 

eigenValues = imagePca.getEigenValues() 

# took in 3 images 

self.assertEqual(len(eigenImages), nCell * nCell) 

self.assertEqual(len(eigenValues), nCell * nCell) 

# all the same shape, only 1 eigenvalue 

self.assertAlmostEqual(eigenValues[0], 1.0) 

self.assertAlmostEqual(eigenValues[1], 0.0) 

self.assertAlmostEqual(eigenValues[2], 0.0) 

##### 

class TestMemory(lsst.utils.tests.MemoryTestCase): 

pass 

def setup_module(module): 

lsst.utils.tests.init() 

254 ↛ 255line 254 didn't jump to line 255, because the condition on line 254 was never trueif __name__ == "__main__": 

lsst.utils.tests.init() 

unittest.main()