Fred Moolekamp added a comment - 23/Aug/16 4:51 PM The skipping worked fine, but when I run all of the tests together testDipole fails in line 293: AssertionError: 1.0 != 1.008598744756823 within 2 places. Any ideas why the rounding would be less accurate in an aggregate test but when running it as a single test? Full error: = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = FAILURES = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = _____________________ DipoleAlgorithmTest.testPsfDipoleFit _____________________   self = <testDipole.DipoleAlgorithmTest testMethod = testPsfDipoleFit> scaling = 100.0   def testPsfDipoleFit( self , scaling = 100. ): psf, psfSum, exposure, s = createDipole( self .w, self .h, self .xc, self .yc, scaling = scaling) source = self .measureDipole(s, exposure) # Recreate the simultaneous joint Psf fit in python fp = source.getFootprint() peaks = fp.getPeaks() speaks = [(p.getPeakValue(), p) for p in peaks] speaks.sort() dpeaks = [speaks[ 0 ][ 1 ], speaks[ - 1 ][ 1 ]] negCenter = afwGeom.Point2D(dpeaks[ 0 ].getFx(), dpeaks[ 0 ].getFy()) posCenter = afwGeom.Point2D(dpeaks[ 1 ].getFx(), dpeaks[ 1 ].getFy()) fneg, negPsfSum, fpos, posPsfSum, residIm = self ._makeModel(exposure, psf, fp, negCenter, posCenter) # Should be close to the same as the inputs; as fracOffset # gets smaller this will be worse. This works for scaling = # 100. self .assertAlmostEqual( 1e - 2 * scaling, - 1e - 2 * fneg, 2 ) > self .assertAlmostEqual( 1e - 2 * scaling, 1e - 2 * fpos, 2 ) E AssertionError: 1.0 ! = 1.008598744756823 within 2 places   tests / testDipole.py: 293 : AssertionError

John Parejko added a comment - 23/Aug/16 4:58 PM Strange. Can you put in a print to see the individual values when you run it yourself, to compare with the aggregate failure?

Fred Moolekamp added a comment - 23/Aug/16 5:18 PM It looks like it might have something to do with how numpy seeds are calculated. I forgot about your comment to put the numpy seed inside the setup_module and now that I did that it works. Before I made the change the values of fpos and fneg were consistently different depending on whether or not a single test was run.

John Parejko added a comment - 23/Aug/16 5:20 PM Never forget my comments! Yes, that right there is exactly why we pre-seed with a specific value inside of setUp.

Fred Moolekamp added a comment - 23/Aug/16 5:21 PM