def test_sampleset_initial_states(self):
bqm = dimod.BinaryQuadraticModel.from_ising({}, {'ab': 1, 'bc': 1, 'ca': 1})
initial_states = dimod.SampleSet.from_samples_bqm({'a': 1, 'b': -1, 'c': 1}, bqm)
response = Neal().sample(bqm, initial_states=initial_states, num_reads=1)
self.assertEqual(len(response), 1)
self.assertEqual(response.first.energy, -1)
def test_sample_qubo(self):
Q = {(0, 1): 1}
resp = Neal().sample_qubo(Q)
row, col = resp.record.sample.shape
self.assertEqual(col, 2) # should get back two variables
self.assertIs(resp.vartype, dimod.BINARY) # should be qubo
def test_geometric_schedule(self):
sampler = Neal()
num_vars = 40
h = {v: -1 for v in range(num_vars)}
J = {(u, v): -1 for u in range(num_vars) for v in range(u, num_vars) if u != v}
num_reads = 10
resp = sampler.sample_ising(h, J, num_reads=num_reads, beta_schedule_type='geometric')
row, col = resp.record.sample.shape
self.assertEqual(row, num_reads)
self.assertEqual(col, num_vars) # should get back two variables
self.assertIs(resp.vartype, dimod.SPIN) # should be ising
with self.assertRaises(ValueError):
sampler.sample_ising(h, J, num_reads=num_reads, beta_schedule_type='asd')
def test_interrupt_error(self):
sampler = Neal()
num_vars = 40
h = {v: -1 for v in range(num_vars)}
J = {(u, v): -1
for u in range(num_vars) for v in range(u, num_vars) if u != v}
num_reads = 100
def f():
raise NotImplementedError
resp = sampler.sample_ising(h,
J,
num_reads=num_reads,
interrupt_function=f)
self.assertEqual(len(resp), 1)
def test_seed(self):
sampler = Neal()
num_vars = 40
h = {v: -1 for v in range(num_vars)}
J = {(u, v): -1
for u in range(num_vars) for v in range(u, num_vars) if u != v}
num_reads = 1000
# test seed exceptions
for bad_seed in (3.5, float("inf"), "string", [], {}):
self.assertRaises(TypeError,
sampler.sample_ising, {}, {},
seed=bad_seed)
for bad_seed in (-1, -100, 2**65):
self.assertRaises(ValueError,
sampler.sample_ising, {}, {},
seed=bad_seed)
# no need to do a bunch of sweeps, in fact the less we do the more
# sure we can be that the same seed is returning the same result
all_samples = []
for seed in (1, 25, 2352, 736145, 5682453):
response0 = sampler.sample_ising(h,
J,
num_reads=num_reads,
num_sweeps=10,
seed=seed)
response1 = sampler.sample_ising(h,
J,
num_reads=num_reads,
num_sweeps=10,
seed=seed)
samples0 = response0.record.sample
samples1 = response1.record.sample
self.assertTrue(np.array_equal(samples0, samples1),
"Same seed returned different results")
for previous_sample in all_samples:
self.assertFalse(np.array_equal(samples0, previous_sample),
"Different seed returned same results")
all_samples.append(samples0)
def test_sample_ising(self):
h = {'a': 0, 'b': -1}
J = {('a', 'b'): -1}
resp = Neal().sample_ising(h, J)
row, col = resp.record.sample.shape
self.assertEqual(col, 2) # should get back two variables
self.assertIs(resp.vartype, dimod.SPIN) # should be ising
def test_one_edge_beta_range(self):
J = {('a', 'b'): 1}
bqm = dimod.BinaryQuadraticModel({}, J, 0, dimod.BINARY)
response = Neal().sample(bqm)
hot_beta, cold_beta = response.info['beta_range']
# Check beta values
# Note: beta is proportional to 1/temperature, therefore hot_beta < cold_beta
self.assertLess(hot_beta, cold_beta)
self.assertNotEqual(hot_beta, float("inf"), "Starting value of 'beta_range' is infinite")
self.assertNotEqual(cold_beta, float("inf"), "Final value of 'beta_range' is infinite")
def test_disconnected_problem(self):
sampler = Neal()
h = {}
J = {
# K_3
(0, 1): -1,
(1, 2): -1,
(0, 2): -1,
# disonnected K_3
(3, 4): -1,
(4, 5): -1,
(3, 5): -1,
}
resp = sampler.sample_ising(h, J, sweeps=1000, num_reads=100)
row, col = resp.record.sample.shape
self.assertEqual(row, 100)
self.assertEqual(col, 6) # should get back two variables
self.assertIs(resp.vartype, dimod.SPIN) # should be ising
def test_cubic_lattice_with_geometric(self):
# Set up all lattice edges in a cube. Each edge is labelled by a 3-D coordinate system
def get_cubic_lattice_edges(N):
for x, y, z in itertools.product(range(N), repeat=3):
u = x, y, z
yield u, ((x+1)%N, y, z)
yield u, (x, (y+1)%N, z)
yield u, (x, y, (z+1)%N)
# Add a J-bias to each edge
np_rand = np.random.RandomState(128)
J = {e: np_rand.choice((-1, 1)) for e in get_cubic_lattice_edges(12)}
# Solve ising problem
sampler = Neal()
response = sampler.sample_ising({}, J, beta_schedule_type="geometric", num_reads=10)
_, response_energy, _ = next(response.data())
# Note: lowest energy found was -3088 with a different benchmarking tool
threshold = -3000
self.assertLess(response_energy, threshold, ("response_energy, {}, exceeds "
"threshold").format(response_energy))
def test_ising_tuple_initial_states(self):
sampler = Neal()
var_labels = ["a", "b", "c", "d"]
num_vars = len(var_labels)
h = {v: -1 for v in var_labels}
J = {(u, v): 1 for u, v in itertools.combinations(var_labels, 2)}
num_reads = 100
seed = 1234567890
np_rand = np.random.RandomState(seed)
initial_state_array = 2*np_rand.randint(2, size=(num_reads, num_vars)) - 1
init_labels = dict(zip(var_labels, np_rand.permutation(num_vars)))
# sample, but ignore deprecation warnings (tuple format for initial_states)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
resp = sampler.sample_ising(h, J, num_reads=num_reads, num_sweeps=0, seed=seed,
initial_states=(initial_state_array, init_labels))
for v in var_labels:
self.assertTrue(np.array_equal(resp.record.sample[:, resp.variables.index(v)],
initial_state_array[:, init_labels[v]]),
"Samples were not the same as initial states with "
"no sweeps")
def test_binary_tuple_initial_states(self):
bqm = dimod.BinaryQuadraticModel.from_ising({}, {(0,1): 1}).binary
num_vars = len(bqm)
num_reads = 10
initial_states_array = np.random.randint(2, size=(num_reads, num_vars))
expected_response = dimod.SampleSet.from_samples_bqm(initial_states_array, bqm)
# sample, but ignore deprecation warnings (tuple format for initial_states)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
response = Neal().sample(bqm, num_sweeps=0, initial_states=(initial_states_array, None))
# if initial_states are not properly converted by the sample method
# (from binary to ising), energy levels will not match
self.assertTrue(np.array_equal(response.record.sample, expected_response.record.sample))
self.assertTrue(np.array_equal(response.record.energy, expected_response.record.energy))
def test_soft_num_reads(self):
"""Number of reads adapts to initial_states size, if provided."""
bqm = dimod.BinaryQuadraticModel.from_ising({}, {'ab': -1, 'bc': 1, 'ac': 1})
init = dimod.SampleSet.from_samples_bqm([{'a': 1, 'b': 1, 'c': 1},
{'a': -1, 'b': -1, 'c': -1}], bqm)
sampler = Neal()
# default num_reads == 1
self.assertEqual(len(sampler.sample(bqm)), 1)
self.assertEqual(len(sampler.sample(bqm, initial_states_generator="random")), 1)
# with initial_states, num_reads == len(initial_states)
self.assertEqual(len(sampler.sample(bqm, initial_states=init)), 2)
# ... but explicit truncation works too
self.assertEqual(len(sampler.sample(bqm, initial_states=init, num_reads=1)), 1)
# if num_reads explicitly given together with initial_states, they are expanded
self.assertEqual(len(sampler.sample(bqm, initial_states=init, num_reads=3)), 3)
# if num_reads explicitly given together without initial_states, they are generated
self.assertEqual(len(sampler.sample(bqm, num_reads=4)), 4)
def test_num_reads(self):
sampler = Neal()
h = {}
J = {('a', 'b'): .5, (0, 'a'): -1, (1, 'b'): 0.0}
for num_reads in (1, 10, 100, 3223, 10352):
response = sampler.sample_ising(h, J, num_reads=num_reads)
row, col = response.record.sample.shape
self.assertEqual(row, num_reads)
self.assertEqual(col, 4)
for bad_num_reads in (0, -1, -100):
with self.assertRaises(ValueError):
sampler.sample_ising(h, J, num_reads=bad_num_reads)
for bad_num_reads in (3.5, float("inf"), "string", [], {}):
with self.assertRaises(TypeError):
sampler.sample_ising(h, J, num_reads=bad_num_reads)
def test_instantiation(self):
sampler = Neal()
dimod.testing.assert_sampler_api(sampler)
def test_job_shop_scheduling_with_linear(self):
# Set up a job shop scheduling BQM
#
# Provide hardcode version of the bqm of "jobs"
# jobs = {'b': [(1,1), (3,1)],
# 'o': [(2,2), (4,1)],
# 'g': [(1,2)]}
#
# There are three jobs: 'b', 'o', 'g'
# Each tuple represents a task that runs on a particular machine for a given amount of
# time. I.e. (machine_id, duration_on_machine)
#
# Variables below are labelled as '<job_name>_<task_index>,<task_start_time>'.
linear = {
'b_0,0': -2.0,
'b_0,1': -2.0,
'b_0,2': -2.0,
'b_0,3': -2.0,
'b_1,0': 0.125,
'b_1,1': -1.5,
'b_1,2': 0.0,
'g_0,0': -1.875,
'g_0,1': -1.5,
'g_0,2': 0.0,
'o_0,0': -2.0,
'o_0,1': -2.0,
'o_0,2': -2.0,
'o_1,0': 0.03125,
'o_1,1': -1.875,
'o_1,2': -1.5,
'o_1,3': 0.0
}
quadratic = {
('b_0,0', 'g_0,0'): 4,
('b_0,1', 'b_0,0'): 4.0,
('b_0,1', 'g_0,0'): 2,
('b_0,2', 'b_0,0'): 4.0,
('b_0,2', 'b_0,1'): 4.0,
('b_0,2', 'b_1,2'): 2,
('b_0,2', 'g_0,1'): 2,
('b_0,2', 'g_0,2'): 4,
('b_0,3', 'b_0,0'): 4.0,
('b_0,3', 'b_0,1'): 4.0,
('b_0,3', 'b_0,2'): 4.0,
('b_0,3', 'b_1,2'): 2,
('b_0,3', 'g_0,2'): 2,
('b_1,1', 'b_0,1'): 2,
('b_1,1', 'b_0,2'): 2,
('b_1,1', 'b_0,3'): 2,
('b_1,1', 'b_1,2'): 4.0,
('g_0,1', 'b_0,1'): 4,
('g_0,1', 'g_0,0'): 4.0,
('g_0,2', 'g_0,0'): 4.0,
('g_0,2', 'g_0,1'): 4.0,
('o_0,0', 'o_1,1'): 2,
('o_0,1', 'o_0,0'): 4.0,
('o_0,1', 'o_1,1'): 2,
('o_0,1', 'o_1,2'): 2,
('o_0,2', 'o_0,0'): 4.0,
('o_0,2', 'o_0,1'): 4.0,
('o_0,2', 'o_1,1'): 2,
('o_1,2', 'o_0,2'): 2,
('o_1,2', 'o_1,1'): 4.0,
('o_1,3', 'o_0,2'): 2,
('o_1,3', 'o_1,1'): 4.0,
('o_1,3', 'o_1,2'): 4.0
}
jss_bqm = dimod.BinaryQuadraticModel(linear, quadratic, 9.0,
dimod.BINARY)
# Optimal energy
optimal_solution = {
'b_0,0': 1,
'b_0,1': 0,
'b_0,2': 0,
'b_0,3': 0,
'b_1,0': 0,
'b_1,1': 1,
'b_1,2': 0,
'b_1,3': 0,
'g_0,0': 0,
'g_0,1': 1,
'g_0,2': 0,
'g_0,3': 0,
'o_0,0': 1,
'o_0,1': 0,
'o_0,2': 0,
'o_0,3': 0,
'o_1,0': 0,
'o_1,1': 0,
'o_1,2': 1,
'o_1,3': 0
}
optimal_energy = jss_bqm.energy(optimal_solution) # Evaluates to 0.5
# Get heuristic solution
sampler = Neal()
response = sampler.sample(jss_bqm, beta_schedule_type="linear")
_, response_energy, _ = next(response.data())
# Compare energies
threshold = 0.1 # Arbitrary threshold
self.assertLess(response_energy, optimal_energy + threshold)
def test_initial_states_generator(self):
bqm = dimod.BinaryQuadraticModel.from_ising({}, {'ab': -1, 'bc': 1, 'ac': 1})
init = dimod.SampleSet.from_samples_bqm([{'a': 1, 'b': 1, 'c': 1},
{'a': -1, 'b': -1, 'c': -1}], bqm)
sampler = Neal()
# 2 fixed initial state, 8 random
resp = sampler.sample(bqm, initial_states=init, num_reads=10)
self.assertEqual(len(resp), 10)
# 2 fixed initial states, 8 random, explicit
resp = sampler.sample(bqm, initial_states=init, initial_states_generator='random', num_reads=10)
self.assertEqual(len(resp), 10)
# all random
resp = sampler.sample(bqm, initial_states_generator='random', num_reads=10)
self.assertEqual(len(resp), 10)
# all random
resp = sampler.sample(bqm, num_reads=10)
self.assertEqual(len(resp), 10)
# zero-length init states in tuple format, extended by random samples
zero_init_tuple = (np.empty((0, 3)), None)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
resp = sampler.sample(bqm, initial_states=zero_init_tuple, num_reads=10)
self.assertEqual(len(resp), 10)
# explicit None for initial_states should use one random init state
resp = sampler.sample(bqm, initial_states=None)
self.assertEqual(len(resp), 1)
# initial_states truncated to num_reads?
resp = sampler.sample(bqm, initial_states=init, initial_states_generator='none', num_reads=1)
self.assertEqual(len(resp), 1)
resp = sampler.sample(bqm, initial_states=init, initial_states_generator='tile', num_reads=1)
self.assertEqual(len(resp), 1)
resp = sampler.sample(bqm, initial_states=init, initial_states_generator='random', num_reads=1)
self.assertEqual(len(resp), 1)
# 2 fixed initial states, repeated 5 times
resp = sampler.sample(bqm, initial_states=init, initial_states_generator='tile', num_reads=10)
self.assertEqual(len(resp), 10)
# can't tile empty states
with self.assertRaises(ValueError):
resp = sampler.sample(bqm, initial_states_generator='tile', num_reads=10)
# not enough initial states
with self.assertRaises(ValueError):
resp = sampler.sample(bqm, initial_states_generator='none', num_reads=3)
# initial_states incompatible with the bqm
init = dimod.SampleSet.from_samples({'a': 1, 'b': 1}, vartype='SPIN', energy=0)
with self.assertRaises(ValueError):
resp = sampler.sample(bqm, initial_states=init)
