def test_line():
model = line_potts_model(gr_size=40, al_size=3, seed=0)
gt = model.infer(algorithm='path_dp')
result = model.infer(algorithm='mean_field')
assert_results_close(result, gt, log_pf_tol=3.0, mp_mse_tol=1e-2)
def test_long_line_compare_with_tree():
# Test to ensure Tree DP and Path DP are consistent.
gr_size = 1000
al_size = 3
model = line_potts_model(gr_size=gr_size, al_size=al_size, seed=111)
result1 = model.infer(algorithm='path_dp')
result2 = model.infer(algorithm='tree_dp')
assert_results_close(result1, result2)
def test_line_potts_4x3_sherali_adams():
"""
Sherali-Adams is exact on line graph.
"""
model = line_potts_model(gr_size=4, al_size=3, seed=0)
max_lh_gt = model.max_likelihood(algorithm='tree_dp')
sa_res = sherali_adams(model, level=3)
max_lh_ub = sa_res.upper_bound
assert np.allclose(max_lh_ub, np.log(model.evaluate(max_lh_gt)), atol=1e-2)
def test_long_line():
gr_size = 1000
al_size = 5
j = np.ones((al_size, al_size)) + np.eye(al_size)
model = line_potts_model(gr_size=gr_size,
al_size=al_size,
seed=111,
same_j=j,
zero_field=True)
result = model.infer(algorithm='tree_dp')
assert np.allclose(result.marg_prob, np.ones((gr_size, al_size)) / al_size)
def test_line_potts_4x2_lasserre():
"""
Second step of Lasserre hierarchy is exact on line graph.
This test is skipped for Windows since the performance of SCS
solver was unstable.
"""
model = line_potts_model(gr_size=4, al_size=2, seed=0)
max_lh_gt = model.max_likelihood(algorithm='tree_dp')
lasserre_res = lasserre(model, level=2)
max_lh_ub = lasserre_res.upper_bound
x = np.log(model.evaluate(max_lh_gt))
assert np.allclose(max_lh_ub, x, atol=1e-2)
def test_antiferromagnetic_ising_line():
# We create Ising model on a line with very high interactions forcing
# samples to be alternate (i.e. 10101010). Then we assert that in at
# least 99% of cases we got alternating sample.
model = line_potts_model(gr_size=100,
al_size=2,
same_j=[[0, 100], [100, 0]],
zero_field=True)
samples = model.sample(num_samples=100)
def is_alternating(x):
return np.all(np.roll(x, 1) == 1 - x)
good_count = sum([is_alternating(state) for state in samples])
assert good_count >= 99
def test_line_1000x10():
model = line_potts_model(gr_size=1000, al_size=10, seed=0)
max_lh_gt = model.max_likelihood(algorithm='tree_dp')
max_lh = max_lh_path_dp(model)
assert np.allclose(max_lh, max_lh_gt)