def test_non_convex(self):
""" Test for non convex cases.
"""
for size in range(2, 101):
cond = random.randint(1, 100)
numpy.random.seed(cond)
res = helpers.gen_general_obj(size, False, cond, 1)
self.assertEqual(res.shape, (size, size))
if size > 50:
self.assertLess(helpers.mindiag(res), 0)
self.assertAlmostEqual(numpy.linalg.cond(res), cond)
numpy.random.seed(cond)
half = helpers.gen_general_obj(size, False, cond, 0.5)
self.assertTrue(np.allclose(res, half * 2))
def test_gen_obj(self):
# A very large matrix generated in a general way is statistically
# very unlikely to be PSD by accident, so we want to observe that
# the generator with convex option makes very large PSD matrices
# and the generator with nonconvex option will nor 'accidentally'
# make PSD matrices (though there is always some chance of a fluke)
# Generate a convex Q for 200 variables with condition number 20
Q = hp.gen_general_obj(200, True, 20, 20)
self.assertAlmostEqual(np.linalg.cond(Q), 20)
self.assertTrue(mh.isPSD(Q))
np.testing.assert_almost_equal(Q, 0.5 * (Q + Q.T))
# Generate a nonconvex Q for 200 variables with condition number 20
Q = hp.gen_general_obj(200, False, 20, 20)
self.assertAlmostEqual(np.linalg.cond(Q), 20)
self.assertFalse(mh.isPSD(Q))
np.testing.assert_almost_equal(Q, 0.5 * (Q + Q.T))