def test_largerSchedule(self):
jobs = {
'small1': [(1, 1)],
'small2': [(2, 2)],
'longJob': [(0, 1), (1, 1), (2, 1)]
}
max_time = 4
# Get exact sample from Job Shop Scheduler BQM
jss = JobShopScheduler(jobs, max_time)
bqm = jss.get_bqm()
response = ExactSolver().sample(bqm)
response_sample = next(response.samples())
# Verify that response_sample obeys constraints
self.assertTrue(jss.csp.check(response_sample))
# Create expected solution
expected = {
"small1_0,0": 1,
"small2_0,0": 1,
"longJob_0,0": 1,
"longJob_1,1": 1,
"longJob_2,2": 1
}
# Compare variable values
self.compare(response_sample, expected)
def test_simple_schedule_more_machines(self):
jobs = {"j0": [(0, 1)], "j1": [(1, 1)], "j2": [(2, 1)]}
max_time = 3
# Get JSS BQM
scheduler = JobShopScheduler(jobs, max_time)
bqm = scheduler.get_bqm()
# Expected solution
expected = {"j0_0,0": 1, "j1_0,0": 1, "j2_0,0": 1}
fill_with_zeros(expected, jobs, max_time)
expected_energy = get_energy(expected, bqm)
# Sampled solution
# response = EmbeddingComposite(DWaveSampler()).sample(bqm, num_reads=10000)
# response_sample, sample_energy, _, chain_break_fraction = next(response.data())
# print("Chain Break Fraction: ", chain_break_fraction)
# response = SimulatedAnnealingSampler().sample(bqm, num_reads=2000, beta_range=[0.01, 10])
response = ExactSolver().sample(bqm)
response_sample, sample_energy, _ = next(response.data())
# Print response
self.assertTrue(scheduler.csp.check(response_sample))
self.assertEqual(expected_energy, sample_energy)
self.compare(response_sample, expected)
def test_denseSchedule(self):
jobs = {
"a": [(1, 2), (2, 2), (3, 2)],
"b": [(3, 3), (2, 1), (1, 1)],
"c": [(2, 2), (1, 3), (2, 1)]
}
max_time = 6
jss = JobShopScheduler(jobs, max_time)
jss.get_bqm() # Run job shop scheduling constraints
# Solution that obeys all constraints
good_solution = {
"a_0,0": 1,
"a_1,2": 1,
"a_2,4": 1,
"b_0,0": 1,
"b_1,4": 1,
"b_2,5": 1,
"c_0,0": 1,
"c_1,2": 1,
"c_2,5": 1
}
fill_with_zeros(good_solution, jobs, max_time)
# Tasks a_1 and b_1 overlap in time on machine 2
overlap_solution = {
"a_0,0": 1,
"a_1,2": 1,
"a_2,4": 1,
"b_0,0": 1,
"b_1,3": 1,
"b_2,5": 1,
"c_0,0": 1,
"c_1,2": 1,
"c_2,5": 1
}
fill_with_zeros(overlap_solution, jobs, max_time)
self.assertTrue(jss.csp.check(good_solution))
self.assertFalse(jss.csp.check(overlap_solution))
def test_relaxedSchedule(self):
jobs = {"breakfast": [("cook", 2), ("eat", 1)], "music": [("play", 2)]}
max_time = 7
jss = JobShopScheduler(jobs, max_time)
jss.get_bqm() # Run job shop scheduling constraints
# Solution obeys all constraints
good_solution = {
"breakfast_0,0": 1,
"breakfast_1,4": 1,
"music_0,3": 1
}
fill_with_zeros(good_solution, jobs, max_time)
# 'breakfast' tasks are out of order
bad_order_solution = {
"breakfast_0,1": 1,
"breakfast_1,0": 1,
"music_0,0": 1
}
fill_with_zeros(bad_order_solution, jobs, max_time)
self.assertTrue(jss.csp.check(good_solution))
self.assertFalse(jss.csp.check(bad_order_solution))
def test_tinySchedule(self):
jobs = {"a": [(1, 1), (2, 1)], "b": [(2, 1)]}
max_time = 2
# Get exact sample from Job Shop Scheduler BQM
jss = JobShopScheduler(jobs, max_time)
bqm = jss.get_bqm()
response = ExactSolver().sample(bqm)
response_sample = next(response.samples())
# Verify that response_sample obeys constraints
self.assertTrue(jss.csp.check(response_sample))
# Create expected solution
expected = {"a_0,0": 1, "a_1,1": 1, "b_0,0": 1}
# Compare variable values
self.compare(response_sample, expected)
def test_stitch_kwargs(self):
"""Ensure stitch_kwargs is being passed to dwavebinarycsp.stitch
"""
jobs = {
"sandwich": [("bread", 1), ("roast_beef", 1)],
"french_toast": [("egg", 1), ("bread", 1)]
}
max_time = 3
# Verify that reasonable stitch args result in a BQM
good_stitch_kwargs = {"max_graph_size": 6, "min_classical_gap": 1.5}
scheduler = JobShopScheduler(jobs, max_time)
bqm = scheduler.get_bqm(good_stitch_kwargs)
self.assertIsInstance(bqm, BinaryQuadraticModel)
# ImpossibleBQM should be raised, as the max_graph size is too small
bad_stitch_kwargs = {"max_graph_size": 0}
scheduler = JobShopScheduler(jobs, max_time)
self.assertRaises(ImpossibleBQM, scheduler.get_bqm, bad_stitch_kwargs)
def test_dense_schedule(self):
# jobs = {'small1': [(1, 1), (0, 2)],
# 'small2': [(2, 2), (0, 1)],
# 'longJob': [(0, 1), (1, 1), (2, 1)]}
# max_time = 4
jobs = {
"j0": [(1, 2), (2, 2), (3, 2)],
"j1": [(3, 4), (2, 1), (1, 1)],
"j2": [(2, 2), (1, 3), (2, 1)]
}
max_time = 7
# Get JSS BQM
scheduler = JobShopScheduler(jobs, max_time)
bqm = scheduler.get_bqm()
# Expected solution
expected = {
"j0_0,0": 1,
"j0_1,2": 1,
"j0_2,4": 1,
"j1_0,0": 1,
"j1_1,4": 1,
"j1_2,5": 1,
"j2_0,0": 1,
"j2_1,2": 1,
"j2_2,5": 1
}
fill_with_zeros(expected, jobs, max_time)
expected_energy = get_energy(expected, bqm)
# Sampled solution
# response = EmbeddingComposite(DWaveSampler()).sample(bqm, num_reads=10000)
# response_sample, sample_energy, _, _ = next(response.data())
# response = SimulatedAnnealingSampler().sample(bqm, num_reads=2000, beta_range=[0.01, 10])
response = TabuSampler().sample(bqm, num_reads=2000)
response_sample, sample_energy, _ = next(response.data())
# Check response sample
self.assertTrue(scheduler.csp.check(response_sample))
self.assertEqual(expected_energy, sample_energy)
self.compare(response_sample, expected)
def test_time_dependent_biases(self):
"""Test that the time-dependent biases that encourage short schedules are applied
appropriately
"""
jobs = {"j1": [("m1", 2), ("m2", 2)], "j2": [("m1", 2)]}
# Create mock object for stitch(..) output
linear = {
'j1_0,0': -2.0,
'j1_0,1': -2.0,
'j1_0,2': -2.0,
'j1_1,2': -2.0,
'j1_1,3': -2.0,
'j1_1,4': -2.0,
'j2_0,0': -6.0,
'j2_0,1': -6.0,
'j2_0,2': -6.0,
'j2_0,3': -6.0,
'j2_0,4': -6.0,
'aux0': -8.0
}
quadratic = {
('j1_0,0', 'j1_0,1'): 4.0,
('j1_0,0', 'j1_0,2'): 4.0,
('j1_0,0', 'j2_0,0'): 4,
('j1_0,0', 'j2_0,1'): 2,
('j1_0,1', 'j1_0,2'): 4.0,
('j1_0,1', 'j1_1,2'): 2,
('j1_0,1', 'j2_0,1'): 4,
('j1_0,1', 'j2_0,2'): 2,
('j1_0,1', 'j2_0,0'): 2,
('j1_0,2', 'j1_1,2'): 2,
('j1_0,2', 'j1_1,3'): 2,
('j1_0,2', 'j2_0,2'): 4,
('j1_0,2', 'j2_0,3'): 2,
('j1_0,2', 'j2_0,1'): 2,
('j1_1,2', 'j1_1,3'): 4.0,
('j1_1,2', 'j1_1,4'): 4.0,
('j1_1,3', 'j1_1,4'): 4.0,
('j2_0,3', 'j2_0,2'): 4,
('j2_0,3', 'j2_0,4'): 4.0,
('j2_0,3', 'j2_0,0'): 4.0,
('j2_0,3', 'j2_0,1'): 4,
('j2_0,3', 'aux0'): 4.0,
('j2_0,2', 'j2_0,4'): 4.0,
('j2_0,2', 'j2_0,0'): 4.0,
('j2_0,2', 'j2_0,1'): 4.0,
('j2_0,2', 'aux0'): 4.0,
('j2_0,4', 'j2_0,0'): 4.0,
('j2_0,4', 'j2_0,1'): 4.0,
('j2_0,4', 'aux0'): 4.0,
('j2_0,0', 'j2_0,1'): 4.0,
('j2_0,0', 'aux0'): 4.0,
('j2_0,1', 'aux0'): 4.0
}
vartype = dwavebinarycsp.BINARY
mock_stitched_bqm = BinaryQuadraticModel(linear, quadratic, 14.0,
vartype)
scheduler = JobShopScheduler(jobs)
with patch.object(dwavebinarycsp,
"stitch",
return_value=mock_stitched_bqm):
bqm = scheduler.get_bqm()
# Check linear biases
# Note: I have grouped the tests by job-task and am comparing the linear biases between
# adjacent times. Tasks that finish before or at the lowerbound of the optimal schedule
# are not penalized with an additional bias; hence all these task-times should have the
# same bias.
# For example, the 0th task of job 2 (aka 'j2_0') will have the same linear
# bias for times 0 through 2 because with these start times, the task would complete
# before the optimal schedule lowerbound. (i.e. "j2_0,0", "j2_0,1", "j2_0,2" all have the
# same linear biases). The remaining task-times will have increasing biases with time, in
# this way, the shortest schedule is encouraged.
self.assertEqual(bqm.linear['j2_0,0'], bqm.linear['j2_0,1'])
self.assertEqual(bqm.linear['j2_0,1'], bqm.linear['j2_0,2'])
self.assertLess(bqm.linear['j2_0,2'], bqm.linear['j2_0,3'])
self.assertLess(bqm.linear['j2_0,3'], bqm.linear['j2_0,4'])
self.assertEqual(bqm.linear['j1_0,0'], bqm.linear['j1_0,1'])
self.assertEqual(bqm.linear['j1_0,1'], bqm.linear['j1_0,2'])
self.assertLess(bqm.linear['j1_1,2'], bqm.linear['j1_1,3'])
self.assertLess(bqm.linear['j1_1,3'], bqm.linear['j1_1,4'])
# Check quadratic biases
# Specifically, quadratic biases should not be penalized when encouraging shorter schedules
# Note: Unable to simply compare dicts as BQM's quadraticview may re-order dict tuple-keys;
# hence, we're comparing BQM adjacencies.
bqm_with_unchanged_quadratic = BinaryQuadraticModel({}, quadratic, 0,
vartype)
self.assertEqual(bqm.adj, bqm_with_unchanged_quadratic.adj)