def test_near_zero(self): eps = sys.float_info.epsilon equal = [(0, 0), (eps, 0), (0, -eps), (-eps, eps)] for a, b in equal: self.assertAlmostEqual(a, b) self.assertTrue(msprime.almost_equal(a, b, abs_tol=1e-9)) not_equal = [(0, 0.0000001), (-0.0000001, 0)] for a, b in not_equal: self.assertNotAlmostEqual(a, b) self.assertFalse(msprime.almost_equal(a, b, abs_tol=1e-9))
def test_near_zero(self): eps = sys.float_info.epsilon equal = [(0, 0), (eps, 0), (0, -eps), (-eps, eps)] for a, b in equal: self.assertAlmostEqual(a, b) self.assertTrue( msprime.almost_equal(a, b, abs_tol=1e-9)) not_equal = [(0, 0.0000001), (-0.0000001, 0)] for a, b in not_equal: self.assertNotAlmostEqual(a, b) self.assertFalse( msprime.almost_equal(a, b, abs_tol=1e-9))
def test_defaults(self): eps = sys.float_info.epsilon equal = [(1, 1), (0, 0), (1 + eps, 1), (1, 1 - eps), (10.000000000001, 10.0)] for a, b in equal: self.assertAlmostEqual(a, b) self.assertTrue(msprime.almost_equal(a, b))
def test_defaults(self): eps = sys.float_info.epsilon equal = [ (1, 1), (0, 0), (1 + eps, 1), (1, 1 - eps), (10.000000000001, 10.0)] for a, b in equal: self.assertAlmostEqual(a, b) self.assertTrue(msprime.almost_equal(a, b))