def test_deprecated_sweeps_name(self):
bqm = dimod.BinaryQuadraticModel.from_ising({}, {'ab': 1})
sampler = Neal()
with warnings.catch_warnings():
warnings.simplefilter("error")
with self.assertRaises(DeprecationWarning):
sampler.sample(bqm, sweeps=10)
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_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)