def test_bimodal_cluster_sampling_statistics(self):
bqm = dimod.BQM.from_qubo({'ab': 1, 'bd': 1, 'dc': 1, 'ca': 1})
nodes = sorted(bqm.variables)
s1 = State.from_samples(dict(zip(nodes, [0, 1, 0, 0])), bqm)
s2 = State.from_samples(dict(zip(nodes, [0, 0, 1, 0])), bqm)
exp1 = SampleSet.from_samples_bqm(dict(zip(nodes, [0, 0, 0, 0])), bqm)
exp2 = SampleSet.from_samples_bqm(dict(zip(nodes, [0, 1, 1, 0])), bqm)
icm = IsoenergeticClusterMove(seed=None)
inp = States(s1, s2)
exp = [exp1, exp2]
# split between [exp1, exp2] and [exp2, exp1] as output samples
# should be ~50%
cnt = 0
n = 100
for _ in range(n):
res = icm.run(inp).result()
r1, r2 = pluck(res, 'samples')
# test responses are valid
self.assertIn(r1, exp)
self.assertIn(r2, exp)
# verify total samples energy doesn't change after ICM
self.assertEqual(self.total_energy(inp), self.total_energy(res))
# count responses
if r1 == exp1 and r2 == exp2:
cnt += 1
self.assertLess(cnt, 0.75 * n)
self.assertGreater(cnt, 0.25 * n)
def test_validation(self):
bqm1 = dimod.BinaryQuadraticModel({'a': 1}, {}, 0, dimod.SPIN)
bqm2 = dimod.BinaryQuadraticModel({'b': 1}, {}, 0, dimod.SPIN)
s1 = State.from_sample({'a': +1}, bqm1)
s2 = State.from_sample({'b': -1}, bqm2)
# two input states required
with self.assertRaises(ValueError):
inp = States(s1, s1, s1)
IsoenergeticClusterMove().run(inp).result()
# variables must match
with self.assertRaises(ValueError):
inp = States(s1, s2)
IsoenergeticClusterMove().run(inp).result()
def test_large_sparse(self):
"Total energy is preserved after ICM on random samples over random graph."
# random Erdős-Rényi sparse graph with 100 nodes and 10% density
graph = nx.generators.fast_gnp_random_graph(n=100, p=0.1)
bqm = dimod.generators.uniform(graph=graph, vartype=dimod.SPIN)
nodes = sorted(bqm.variables)
# random input samples
s1 = State.from_problem(bqm, samples=random_sample)
s2 = State.from_problem(bqm, samples=random_sample)
inp = States(s1, s2)
icm = IsoenergeticClusterMove()
res = icm.run(inp).result()
self.assertAlmostEqual(self.total_energy(inp), self.total_energy(res))
def test_ising_triangle_flip(self):
bqm = dimod.BQM.from_ising({}, {'ab': 1, 'bc': 1, 'ca': 1})
s1 = State.from_samples({'a': -1, 'b': +1, 'c': +1}, bqm)
s2 = State.from_samples({'a': +1, 'b': -1, 'c': +1}, bqm)
icm = IsoenergeticClusterMove()
inp = States(s1, s2)
res = icm.run(inp).result()
# Expected: ('a', 'b') identified as (the sole) cluster, selected,
# resulting in variables {'a', 'b'} flipped. Effectively, input states
# are simply swapped.
self.assertEqual(res[0].samples, s2.samples)
self.assertEqual(res[1].samples, s1.samples)
# verify total samples energy doesn't change after ICM
self.assertEqual(self.total_energy(inp), self.total_energy(res))
def test_small_lattice(self):
graph = nx.generators.lattice.grid_2d_graph(5, 5)
bqm = dimod.generators.uniform(graph,
vartype=dimod.BINARY,
low=1,
high=1)
nodes = sorted(bqm.variables)
s1 = State.from_samples(
dict(
zip(nodes, [
0, 1, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 1, 0,
1, 0, 1, 0, 0
])), bqm)
s2 = State.from_samples(
dict(
zip(nodes, [
0, 1, 1, 0, 0, 0, 0, 0, 1, 1, 0, 1, 1, 0, 1, 0, 1, 0, 0, 0,
1, 0, 0, 1, 0
])), bqm)
exp1 = SampleSet.from_samples_bqm(
dict(
zip(nodes, [
0, 1, 0, 1, 1, 0, 0, 0, 1, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0, 0,
1, 0, 0, 1, 0
])), bqm)
exp2 = SampleSet.from_samples_bqm(
dict(
zip(nodes, [
0, 1, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1, 0, 0, 1, 1, 0,
1, 0, 1, 0, 0
])), bqm)
icm = IsoenergeticClusterMove(seed=1234)
inp = States(s1, s2)
res = icm.run(inp).result()
self.assertEqual(res[0].samples, exp1)
self.assertEqual(res[1].samples, exp2)
# verify total samples energy doesn't change after ICM
self.assertEqual(self.total_energy(inp), self.total_energy(res))